[ { "id": "nvidia-blackwell", "name": "Blackwell GPU B200/B300", "company": "NVIDIA", "region": "global", "tagline": "NVIDIA's Blackwell datacenter AI generation — B200 and B300 GPUs, GB200/GB300 NVL72 racks", "description": "Blackwell is NVIDIA's current datacenter AI generation. It is sold as the B200 GPU, the higher-memory B300 (Blackwell Ultra), and as the GB200 and GB300 NVL72 rack-scale systems, which link 72 Blackwell GPUs and 36 Grace CPUs into a single NVLink domain for large-model training and inference. Buyers evaluate the generation, not the individual SKU, so the GPUs and the rack systems are one product row. What is recorded here is the platform surface and its hyperscaler/OEM role; NVIDIA does not disclose per-SKU shipment or deployment counts.", "users": "Hyperscaler, cloud and OEM AI infrastructure; unit and deployment counts not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 94, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [ "B200 and B300 (Blackwell Ultra) datacenter GPUs", "GB200 and GB300 NVL72 rack-scale systems (72 Blackwell GPUs, 36 Grace CPUs, one NVLink domain)", "Large-model training and inference", "CUDA, networking and software-stack integration", "Hyperscaler, cloud-provider and OEM AI-factory deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/", "apiLink": null, "aiCoreShare": 84, "adoptionSignal": "NVIDIA positions Blackwell as a datacenter-scale AI infrastructure generation for hyperscalers, cloud providers, OEM systems, and enterprise AI factories. App-style active users are not the right evidence type for an accelerator row, and NVIDIA does not disclose per-SKU unit counts.", "sourceUrls": [ "https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/", "https://www.nvidia.com/en-us/data-center/gb200-nvl72/" ], "scoringRationale": "Blackwell is NVIDIA-class global AI infrastructure: current datacenter accelerator generation, hyperscaler/OEM system role, high product criticality, broad software-ecosystem dependence, and direct substitution against AMD Instinct, Google TPU, Groq, Cerebras, Rebellions, and FuriosaAI compute. It scores in the frontier infrastructure tier even without app-style user counts. B200 and GB200 NVL72 were separate rows until 2026-07-29; they are the same generation and now share one page.", "versions": [ { "name": "Blackwell Ultra (B300 / GB300 NVL72)", "released": "2025", "variants": [ "B300 GPU", "GB300 NVL72 rack system" ], "note": "Higher-memory refresh of the Blackwell generation." }, { "name": "Blackwell (B200 / GB200 NVL72)", "released": "2024", "variants": [ "B200 GPU", "GB200 NVL72 rack system", "GB200 Grace Blackwell Superchip" ], "note": "Succeeded Hopper; Rubin is the announced successor." } ], "versionsUpdated": "2026-07-29" }, { "id": "nvidia-hopper", "name": "Hopper GPU H100/H200", "company": "NVIDIA", "region": "global", "tagline": "NVIDIA's Hopper datacenter GPU generation — H100 and H200", "description": "Hopper is the NVIDIA datacenter GPU generation behind most current foundation-model training and inference. It ships as the H100 and as the higher-memory H200, across SXM, PCIe and NVL form factors. The generation is what a buyer evaluates and what clouds rent by the hour, so H100 and H200 are one product row. NVIDIA does not disclose unit or customer counts for either SKU.", "users": "Global AI datacenter standard; unit counts not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 93, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [ "H100 Tensor Core GPU", "H200 Tensor Core GPU with higher memory capacity and bandwidth", "SXM, PCIe and NVL form factors", "Foundation-model training and inference", "Hyperscaler, cloud-provider and enterprise GPU-cluster deployment", "Suite modules: H200 Tensor Core GPU", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.nvidia.com/en-us/data-center/technologies/hopper-architecture/", "apiLink": null, "aiCoreShare": 82, "adoptionSignal": "Hopper remains a widely deployed AI training and inference generation across hyperscalers, AI labs, cloud providers, and enterprise GPU clusters. Unit and customer counts are not recorded by NVIDIA at SKU level.", "sourceUrls": [ "https://www.nvidia.com/en-us/data-center/technologies/hopper-architecture/", "https://www.nvidia.com/en-us/data-center/h200/" ], "scoringRationale": "Hopper is one of the canonical global AI infrastructure products behind modern foundation-model training and inference. Infrastructure rows are judged on datacenter role and ecosystem leverage rather than app-style user counts. It sits just below Blackwell because Blackwell is the current shipping generation. H100 and H200 were separate rows until 2026-07-29; they are the same generation and now share one page.", "versions": [ { "name": "Hopper (H100 / H200)", "released": "2022", "variants": [ "H100 SXM", "H100 PCIe", "H100 NVL", "H200" ], "note": "Preceding NVIDIA datacenter generation; still the volume installed base. Succeeded by Blackwell." } ], "versionsUpdated": "2026-07-29" }, { "id": "nvidia-isaac-sim", "name": "NVIDIA Isaac Sim", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://developer.nvidia.com/isaac/sim", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "rebel", "name": "RebelServer", "company": "Rebellions", "region": "korea", "tagline": "Rebellions server for efficient large-scale AI inference", "description": "RebelServer is Rebellions' Rebel100-powered inference server for sovereign and enterprise AI deployments, with 8 RebelCards per 5U server and support for LLM, MoE, multimodal, vision, and speech workloads.", "users": "Enterprise and government environments", "userCount": 0, "adoptionSignal": "Rebellions describes RebelServer as production-proven across enterprise and government environments, but does not disclose named customers or deployment counts.", "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "groq-lpu", "cerebras-inference", "furiosa-rngd", "tenstorrent-ai-cloud" ], "features": [ "8 RebelCards per server", "Up to 2 PFLOPS FP8", "LLM and MoE inference", "vLLM, PyTorch, and Triton integration" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://rebellions.ai/rebellions-product/rebelserver/", "apiLink": null, "aiCoreShare": 83, "scoringRationale": "Capped at 78 under the AI hardware evidence cap: userCount is zero, no named hyperscaler or cloud deployment is disclosed, and no quantified adoption exists. Strong technical specs (2 PFLOPS FP8, 8 RebelCards) are not a substitute for deployment proof. Applied at the same ceiling as inference chips without verified production customers globally. Clears to 84 when a named cloud or OEM deployment is sourced; clears further with quantified adoption.", "sourceUrls": [ "https://rebellions.ai/rebellions-product/rebelserver/", "https://rebellions.ai/" ] }, { "id": "furiosa-rngd", "name": "RNGD PCIe", "company": "FuriosaAI", "region": "korea", "tagline": "FuriosaAI PCIe accelerator for LLM and multimodal inference", "description": "RNGD PCIe is FuriosaAI's second-generation AI accelerator card for LLM and multimodal inference, built on Furiosa's tensor-contraction processor architecture.", "descriptionKo": "RNGD PCIe는 퓨리오사AI의 2세대 AI 가속기 카드로, 자체 개발한 텐서 수축 프로세서(TCP) 아키텍처를 기반으로 LLM 및 멀티모달 추론을 처리합니다. 퓨리오사 소프트웨어 스택과 함께 제공되며, RDNA 기반 경쟁 제품과 달리 처음부터 LLM 추론에 특화된 설계를 채택한 것이 기술적 차별점입니다. 현재 파트너사 로고와 Furiosa Access 거점은 공개되어 있으나, 제품별 고객 수나 출하 수량은 공개되지 않아 AI 하드웨어 증거 기준(상한 78)이 적용됩니다. 명칭이 있는 클라우드 또는 OEM 배포가 확인되면 84점으로 상향될 수 있습니다.", "users": "Enterprise evaluation; product customers not disclosed", "userCount": 0, "releaseDate": "2024-01", "technicalImpact": 78, "competitors": [ "atom-max", "groq-lpu", "cerebras-inference", "nvidia", "amd" ], "features": [ "Tensor Contraction Processor architecture", "LLM inference", "Multimodal inference", "Furiosa software stack" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales", "commercial": "Contact FuriosaAI" }, "link": "https://furiosa.ai/rngd", "apiLink": null, "aiCoreShare": 89, "adoptionSignal": "Furiosa lists named partner logos and worldwide Furiosa Access locations, but does not disclose product-specific customers or shipped units.", "scoringRationale": "Capped at 78 under the AI hardware evidence cap: userCount is zero, no named hyperscaler or cloud deployment is disclosed, and partner logos do not constitute quantified adoption. Tensor Contraction Processor architecture is technically credible, but specs alone are not scoring inputs. Applied at the same ceiling as inference chips without verified production customers globally. Clears to 84 when a named cloud or OEM deployment is sourced; clears further with quantified adoption.", "sourceUrls": [ "https://furiosa.ai/", "https://furiosa.ai/rngd" ] }, { "id": "cann", "name": "CANN", "company": "Huawei", "region": "china", "tagline": "Huawei's compute architecture and toolchain for Ascend silicon", "description": "CANN (Compute Architecture for Neural Networks) is Huawei's software layer between AI frameworks and Ascend accelerators, covering operator libraries, the graph compiler, runtime and profiling tools. It is what makes Ascend hardware usable from MindSpore, PyTorch and TensorFlow, and it is the switching cost that keeps Chinese AI workloads on Huawei silicon. Huawei discloses no developer count or deployment figure.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.hiascend.com/cann", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://www.hiascend.com/cann" ], "scoringRationale": "CANN is Huawei software infrastructure for Ascend deployment. It supports ecosystem control, but remains below hardware/platform rows with stronger adoption evidence." }, { "id": "atom-max", "name": "ATOM-Max", "company": "Rebellions", "region": "korea", "tagline": "Rebellions ATOM inference systems from single server to rack-scale POD", "description": "ATOM-Max is Rebellions' NPU server product line for inference deployment. In August 2026, Rebellions announced ATOM-Max server supply to Gyeongsangnam-do and Ulsan in a public AI CCTV conversion project. This is specific customer evidence, not proof of broad commercial scale. Earlier ATOM-Plus cloud deployments should not be attributed to ATOM-Max.", "descriptionKo": "ATOM-Max는 리벨리온의 추론용 NPU 서버 제품군입니다. 회사는 2026년 8월 공공 AI CCTV 전환 사업을 통해 경상남도와 울산에 ATOM-Max 서버를 공급한다고 밝혔습니다. 실명 고객 근거는 있지만 공급 대수·계약 매출·반복 구매는 공개되지 않았습니다. ATOM-Plus의 클라우드 실적을 ATOM-Max 실적으로 간주하지 않습니다.", "users": "Named Korean public-sector deliveries; unit count not disclosed", "userCount": 0, "adoptionSignal": "Rebellions named Gyeongsangnam-do and Ulsan as recipients, with delivery starting August 2026 in one public CCTV program. Unit counts, contract revenue and repeat purchases are not disclosed.", "releaseDate": "", "technicalImpact": 82, "competitors": [ "furiosa-rngd", "furiosaai-sovereign-ai-appliance", "NVIDIA Hopper", "AMD Instinct", "groq-lpu", "cerebras-inference", "tenstorrent-ai-cloud" ], "features": [ "4U server with up to 8 ATOM-Max PCIe cards", "1,024 TFLOPS FP16 peak and 512GB GDDR6 per server", "8-server Mini POD with 64 NPUs and 400GB/s RDMA fabric", "vLLM, Triton, Kubernetes, Hugging Face, SDK, and model-zoo support" ], "pricing": { "commercial": "Contact Rebellions", "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://rebellions.ai/rebellions-product/atom-max-server/", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://rebellions.ai/rebellions-product/atom-max-server/", "https://kr.rebellions.ai/newsroom/public_procurement_innovation_product/" ], "scoringRationale": "Reviewed 2026-09-29: retains Impact 82 for limited domestic deployment. Two recipients in one procurement program are not independent broad-market traction. Procurement eligibility, program budget and specifications do not raise Impact; ATOM-Plus evidence is excluded." }, { "id": "coral-ai", "name": "Coral AI", "company": "Google Cloud", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://cloud.google.com/tpu", "apiLink": null, "aiCoreShare": 83, "contentReview": "needs-review" }, { "id": "ascend-910", "name": "Ascend 910B/910C", "company": "Huawei", "region": "china", "tagline": "Huawei's Ascend 910 AI accelerator series — 910B and 910C", "description": "Ascend 910 is Huawei's datacenter AI accelerator series and the primary domestic alternative to NVIDIA compute in China, paired with the CANN software stack and MindSpore. The 910B and 910C are steps in one series rather than separate products. Huawei does not disclose shipment or customer counts, and export-control conditions make third-party verification difficult.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Ascend 910B" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://e.huawei.com/en/products/computing/ascend", "apiLink": null, "aiCoreShare": 87, "sourceUrls": [ "https://e.huawei.com/en/products/computing/ascend" ], "scoringRationale": "Reduced from 84 to 78 by the tiered AI-hardware evidence cap. The row records no named cloud or OEM deployment, no shipment volume and no customer count, which is tier 0 — the same position as any other accelerator with an unproven deployment story. Strategic importance to China's AI stack is not a scoring input; technical specs are not either. Clears to 82 with one named cloud or OEM deployment, and to 86 with multi-partner reach or a disclosed shipment volume.", "versions": [ { "name": "Ascend 910C", "released": "2025", "variants": [ "Ascend 910B", "Ascend 910C" ], "note": "Current series step. Deployment volumes are not publicly disclosed." } ], "adoptionSignal": "Ascend 910B and 910C are Huawei's datacenter AI accelerators and the primary domestic alternative to NVIDIA compute in China, paired with the CANN stack and MindSpore. Huawei discloses no shipment volumes, named cloud or OEM deployments, or customer counts, and export-control conditions make independent verification difficult.", "versionsUpdated": "2026-07-29" }, { "id": "nvidia-isaac", "name": "NVIDIA Isaac", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://nvidia.com", "apiLink": null, "aiCoreShare": 76, "contentReview": "needs-review" }, { "id": "nvidia-jetson-orin", "name": "NVIDIA Jetson Orin", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "Hailo", "Ambarella", "Untether AI", "BrainChip" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://nvidia.com", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "waymo-zeekr-robotaxi", "name": "Waymo Zeekr Robotaxi", "company": "Waymo", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "Tesla FSD", "Mobileye", "Cruise", "Wayve" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://waymo.com", "apiLink": null, "aiCoreShare": 87, "contentReview": "needs-review" }, { "id": "cerebras-inference", "name": "Cerebras Inference", "company": "Cerebras", "region": "global", "tagline": "Wafer-scale inference cloud for low-latency LLM serving", "description": "Cerebras Inference is Cerebras' API and cloud inference service built on Wafer-Scale Engine systems. It is evaluated as global AI infrastructure because Cerebras publishes a self-serve developer path, enterprise inference plans, OpenAI-compatible API access, and named customer or partner stories rather than only a chip datasheet.", "users": "Named enterprise and developer customers; active user count not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 85, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "groq-lpu", "furiosa-rngd", "atom-max", "vLLM", "TensorRT-LLM", "Together AI" ], "features": [ "Wafer-scale inference", "OpenAI-compatible API", "Self-serve and enterprise plans", "Named customer stories", "Suite modules: Cerebras Inference API", "Self-serve API keys", "Enterprise capacity", "Wafer-scale inference backend" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.cerebras.ai/inference", "apiLink": null, "aiCoreShare": 88, "adoptionSignal": "Cerebras publishes a live inference cloud, free-trial/API access, enterprise plans, and named customer or partner stories including GSK, DeepLearning.AI, Meta/Llama, and Perplexity.", "scoringRationale": "Cerebras Inference is the cloud inference platform built on Cerebras's wafer-scale chip architecture, offering the highest single-chip memory bandwidth available and sub-50ms TTFT on frontier models. Named enterprise and developer customers are confirmed. Reduced from 88 to 85: at 88 it exceeded Midjourney (87, 21M verified users), which is not justified for a startup with named but undisclosed customer counts. 85 reflects a technically differentiated inference platform with real customers but without the scale evidence to match top-tier hardware infrastructure rows. Re-evaluates upward if Cerebras discloses customer count, revenue, or API call volume.", "sourceUrls": [ "https://www.cerebras.ai/inference", "https://www.cerebras.ai/chip" ], "versions": [ { "name": "Cerebras Inference", "variants": [ "Cerebras Inference API" ] } ], "versionsUpdated": "2026-07-29" }, { "id": "figure-humanoid-workforce", "name": "Figure Humanoid Workforce", "company": "Figure AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "figure", "1X NEO", "Tesla Optimus", "Boston Dynamics Atlas" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://figure.ai", "apiLink": null, "aiCoreShare": 88, "contentReview": "needs-review" }, { "id": "cloud-tpu-v5", "name": "Cloud TPU v5e/v5p", "company": "Google Cloud", "region": "global", "tagline": "Google Cloud's fifth-generation TPU accelerators — v5e and v5p", "description": "Cloud TPU v5 is Google Cloud's fifth-generation tensor processing unit, rented as v5e for cost-efficient inference and smaller training jobs and as v5p for large-scale training. The two are configurations of one generation rather than separate products, so they share a row. Trillium (v6) is a later generation and stays separate.", "users": "Rentable on Google Cloud worldwide; unit counts not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "AWS Trainium 2" ], "features": [ "Fifth-generation Google Cloud TPU accelerators", "v5e for cost-efficient inference and mid-scale training", "v5p for large-scale model training", "Pod-scale interconnect", "JAX, PyTorch and TensorFlow support", "Suite modules: Cloud TPU v5e", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://cloud.google.com/tpu", "apiLink": null, "aiCoreShare": 76, "versions": [ { "name": "Cloud TPU v5", "released": "2023", "variants": [ "v5e (inference / mid-scale training)", "v5p (large-scale training)" ], "note": "Trillium (TPU v6) is the later generation and is catalogued separately." } ], "sourceUrls": [], "adoptionSignal": "Cloud TPU v5e and v5p are Google's own accelerators, available for rent across Google Cloud regions worldwide and used for Google's internal model training. Google does not disclose deployed unit counts or customer numbers.", "versionsUpdated": "2026-07-29" }, { "id": "edge-tpu", "name": "Edge TPU", "company": "Google Cloud", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://cloud.google.com/tpu", "apiLink": null, "aiCoreShare": 80, "contentReview": "needs-review" }, { "id": "atlas-ai-computing-platform", "name": "Atlas AI Computing Platform", "company": "Huawei", "region": "china", "tagline": "Huawei's Atlas AI compute systems for datacentre and edge", "description": "Atlas is Huawei's line of AI compute hardware — accelerator cards, servers, edge boxes and cluster appliances built on Ascend silicon and programmed through the CANN stack and MindSpore. It is the main domestic alternative for Chinese datacentre and industrial AI buyers facing export-control constraints on foreign accelerators. Huawei does not disclose shipment volumes, named cloud deployments or customer counts for the Atlas line.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 82, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://e.huawei.com/en/products/computing/ascend", "apiLink": null, "aiCoreShare": 85, "sourceUrls": [ "https://e.huawei.com/en/products/computing/ascend" ], "scoringRationale": "Atlas is Huawei enterprise AI-computing infrastructure around Ascend. It has strategic platform relevance, but not enough disclosed deployment evidence for top-tier scoring." }, { "id": "furiosaai-sovereign-ai-appliance", "name": "NXT RNGD Server", "company": "FuriosaAI", "region": "korea", "tagline": "3 kW FuriosaAI inference appliance for agentic systems", "description": "NXT RNGD Server is FuriosaAI's multi-card RNGD inference appliance for LLM and agentic AI workloads, using eight RNGD cards in an air-cooled 3 kW server for on-prem, managed, or colocation deployments.", "users": "Furiosa Access evaluation locations", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "groq-lpu", "cerebras-inference", "atom-max", "tenstorrent-ai-cloud" ], "features": [ "8 RNGD cards", "4 petaFLOPS FP8", "384GB HBM3", "3 kW air-cooled server" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://furiosa.ai/nxt-rndg-server", "apiLink": null, "aiCoreShare": 78, "adoptionSignal": "Furiosa lists Access Program locations in Seoul, Bay Area, Lisbon, and Johor Bahru, plus partner logos; product-specific customer counts are not disclosed.", "scoringRationale": "Capped at 78 under the AI hardware evidence cap: userCount is zero, no named hyperscaler or cloud deployment is disclosed, and evaluation-program locations are not equivalent to production deployments. Identical evidence gap to the standalone RNGD PCIe card (also 78) and RebelServer (also 78). Clears to 84 when a named cloud or OEM deployment is sourced; clears further with quantified adoption.", "sourceUrls": [ "https://furiosa.ai/", "https://furiosa.ai/nxt-rndg-server" ] }, { "id": "mindspore", "name": "MindSpore", "company": "Huawei", "region": "china", "tagline": "Huawei's open-source deep-learning framework for Ascend", "description": "MindSpore is Huawei's open-source training and inference framework, designed to run natively on Ascend accelerators with support for GPU and CPU targets. It anchors the domestic alternative stack alongside CANN and Atlas hardware, and competes with PaddlePaddle for Chinese developer share. Huawei publishes the project openly but discloses no developer or production-deployment count.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.mindspore.cn/en", "apiLink": null, "aiCoreShare": 77, "sourceUrls": [ "https://www.mindspore.cn/en" ], "scoringRationale": "MindSpore is an important Chinese AI framework surface, but the catalog does not show the developer adoption needed for a higher score." }, { "id": "nvidia-omniverse", "name": "NVIDIA Omniverse", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://nvidia.com", "apiLink": null, "aiCoreShare": 88, "contentReview": "needs-review" }, { "id": "trillium-tpu", "name": "Trillium TPU", "company": "Google Cloud", "region": "global", "tagline": "", "description": "", "users": "Rentable on Google Cloud worldwide; unit counts not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://cloud.google.com/tpu", "apiLink": null, "aiCoreShare": 75, "adoptionSignal": "Trillium (TPU v6) is Google's own sixth-generation accelerator, available for rent across Google Cloud regions worldwide. Google does not disclose deployed unit counts or customer numbers.", "contentReview": "needs-review" }, { "id": "nvidia-rubin", "name": "NVIDIA Rubin", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "AMD Instinct", "Cloud TPU v5", "AWS Trainium 2" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://nvidia.com", "apiLink": null, "aiCoreShare": 85, "contentReview": "needs-review" }, { "id": "furiosa-sdk", "name": "Furiosa SDK", "company": "FuriosaAI", "region": "korea", "tagline": "Compiler and runtime for FuriosaAI NPUs", "description": "The Furiosa SDK is the compiler, runtime and profiling toolchain that lets PyTorch and ONNX models run on FuriosaAI's RNGD and Warboy accelerators, with vLLM-compatible serving. It is the layer that decides whether Furiosa silicon is practically adoptable by Korean datacentre buyers. Developer and deployment counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://furiosa.ai/software", "apiLink": null, "aiCoreShare": 81, "sourceUrls": [ "https://furiosa.ai/", "https://furiosa.ai/software" ], "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 80 -> 76. A toolchain row cannot outscore the silicon it serves — RNGD sits at 78 under the AI-hardware evidence cap. Sub-row rule places the SDK below the chip surface." }, { "id": "nvidia-jetson-thor", "name": "NVIDIA Jetson Thor", "company": "NVIDIA", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "Hailo", "Ambarella", "Untether AI", "BrainChip" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://nvidia.com", "apiLink": null, "aiCoreShare": 91, "contentReview": "needs-review" }, { "id": "warboy", "name": "Vision NPU", "company": "FuriosaAI", "region": "korea", "tagline": "FuriosaAI vision accelerator for computer-vision workloads", "description": "Vision NPU is FuriosaAI's computer-vision accelerator product line, now presented separately from the RNGD family on Furiosa's current product navigation.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Computer-vision acceleration", "Edge and server inference", "Furiosa software stack", "Industrial vision workloads" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://furiosa.ai/warboy", "apiLink": null, "aiCoreShare": 89, "sourceUrls": [ "https://furiosa.ai/", "https://furiosa.ai/warboy" ], "scoringRationale": "First-generation Warboy computer-vision chip, now superseded by the RNGD family as FuriosaAI's current flagship silicon. No disclosed production deployments or shipment volume in the catalog. Kept below the RNGD score tier; maintained as an active product line for legacy vision-inference customers." }, { "id": "groq-lpu", "name": "Groq LPU (chip/GroqNode/GroqRack)", "company": "Groq", "region": "global", "tagline": "Groq's LPU inference hardware, from chip to node and rack", "description": "The Groq LPU is a deterministic inference processor sold as the chip, as GroqNode servers, and as GroqRack clusters. Those are packaging tiers of one hardware line rather than three products, so they share a row. GroqCloud, the hosted API built on the same hardware, stays a separate row because it is a different purchase for a different buyer.", "users": "3M developers and teams reported by Groq; active users not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 83, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "cerebras-inference", "furiosa-rngd", "atom-max", "tenstorrent-ai-cloud", "vLLM", "TensorRT-LLM" ], "features": [ "Custom LPU architecture", "Deterministic inference scheduling", "GroqCloud serving", "On-prem GroqRack option", "Suite modules: Groq LPU Inference Engine, GroqRack, GroqNode", "Developer workflow automation", "Model operations", "Observability", "Secure deployment", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://groq.com/lpu-architecture", "apiLink": null, "aiCoreShare": 89, "adoptionSignal": "Groq reports 3M developers and teams for GroqCloud, named customer logos, public/private/co-cloud deployment options, and four global data-center regions.", "scoringRationale": "The Groq LPU (Language Processing Unit) is Groq's custom SIMD inference chip delivering deterministic, high-throughput token generation. It is the hardware foundation of GroqCloud and the LPU Inference Engine. Reduced from 88 to 83 to correct double-counting: GroqCloud (85) already captures the commercial evidence (3M company-reported developers); scoring the underlying chip at 88 counts the same user base twice. 83 reflects real hardware differentiation and demonstrated deployment via GroqCloud without double-inflating Groq's catalog footprint. Re-evaluates upward if Groq discloses chip shipment volumes or OEM/hyperscaler partnerships.", "sourceUrls": [ "https://groq.com/lpu-architecture", "https://groq.com/groqcloud" ], "versions": [ { "name": "Groq LPU", "variants": [ "LPU inference engine", "GroqNode server", "GroqRack cluster" ], "note": "GroqCloud is the hosted API on the same hardware and is catalogued separately." } ], "versionsUpdated": "2026-07-29" }, { "id": "horizon-superdrive", "name": "Horizon SuperDrive", "company": "Horizon Robotics", "region": "china", "tagline": "Horizon Robotics' assisted-driving stack on Journey silicon", "description": "SuperDrive is Horizon Robotics' assisted-driving system — perception, planning and control software plus the TogetherOS automotive operating system that OEMs and Tier-1 suppliers integrate against, all running on Horizon's Journey silicon. Horizon is HKEX-listed and supplies Chinese OEM programmes; SuperDrive-specific vehicle volumes are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Horizon TogetherOS" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://horizon.cc", "apiLink": null, "aiCoreShare": 92, "sourceUrls": [], "versions": [ { "name": "Horizon SuperDrive", "variants": [ "Horizon TogetherOS" ] } ], "scoringRationale": "78 reflects an HKEX-listed automotive supplier shipping assisted-driving silicon and software into Chinese OEM programmes. Held at the cap because SuperDrive-specific vehicle volumes and named production programmes are not disclosed, unlike StradVision's sourced 5M+ production vehicles at 84.", "versionsUpdated": "2026-07-29" }, { "id": "tesla-fsd", "name": "Tesla FSD", "company": "Tesla", "region": "global", "tagline": "Tesla's vision-only autonomous driving stack", "description": "Tesla FSD is the company's driver-assistance and autonomy system, built on a vision-only perception stack with no lidar or radar, trained on fleet video and shipped over the air to customer vehicles. Autopilot is the base tier of the same stack and Tesla Vision is the perception layer beneath both.", "users": "1.1M+ subscribers", "userCount": 1100000, "releaseDate": "2020-10", "technicalImpact": 84, "competitors": [ "waymo", "cruise", "mobileye", "Wayve", "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "End-to-end neural network", "Highway autopilot", "City streets navigation", "Auto parking", "Summon", "Suite modules: Full Self-Driving", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Autopilot, Tesla Vision" ], "pricing": { "subscription": "$99/month", "purchase": "$8,000 one-time" }, "link": "https://tesla.com/autopilot", "apiLink": null, "aiCoreShare": 92, "sourceUrls": [ "https://www.tesla.com/en_US/autopilot", "https://ir.tesla.com/", "https://www.tesla.com/autopilot" ], "scoringRationale": "Tesla Full Self-Driving is the most-deployed consumer autonomous-driving AI system globally, with 1.1M+ FSD subscribers and fleet-scale real-world data collection advantage. Score reflects durable hardware-software integration, real-world miles data flywheel, and subscription revenue. Kept at 84 rather than higher because full Level 4/5 autonomy is not yet achieved and regulatory approval for unsupervised use remains limited.", "versions": [ { "name": "Tesla FSD", "released": "2020-10", "variants": [ "Autopilot", "Tesla Vision" ] } ], "versionsUpdated": "2026-07-29" }, { "id": "cerebras-cs-3", "name": "Cerebras CS-3 (WSE-3)", "company": "Cerebras", "region": "global", "tagline": "Cerebras wafer-scale AI system, from WSE-3 chip to Condor Galaxy clusters", "description": "CS-3 is Cerebras's wafer-scale AI system: the WSE-3 processor, the CS-3 system built around it, the wafer-scale cluster configuration, and the Condor Galaxy supercomputers assembled from them. Those are one hardware line at different assembly levels rather than four products. Cerebras Inference, the hosted API on the same hardware, stays a separate row.", "users": "Enterprise and research customers; system shipment count not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 84, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "groq-lpu", "furiosa-rngd", "atom-max" ], "features": [ "Wafer-scale compute system", "WSE-3 based architecture", "Training and inference workloads", "Enterprise AI infrastructure", "Suite modules: Wafer-Scale Engine 3, Cerebras Wafer-Scale Cluster, Cerebras Condor Galaxy", "Wafer-scale processor", "4 trillion transistors", "AI training and inference", "CS-3 system foundation", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.cerebras.ai/chip", "apiLink": null, "aiCoreShare": 90, "adoptionSignal": "Cerebras publishes WSE-3 system claims plus inference-cloud customer stories; system-level shipment and customer counts are not disclosed.", "scoringRationale": "Cerebras CS-3 is the third-generation wafer-scale AI training system — the world's largest chip by die area. Strong academic and enterprise research adoption. Reduced from 86 to 83: hardware scored below Cerebras Inference (85), which has broader buyer access via cloud API. No unit shipment counts or revenue are disclosed. 83 reflects real hardware differentiation for large-model training workloads without the commercial scale evidence to sit above the cloud inference row.", "sourceUrls": [ "https://www.cerebras.ai/chip", "https://www.cerebras.ai/inference" ], "versions": [ { "name": "CS-3 (WSE-3)", "variants": [ "WSE-3 processor", "CS-3 system", "Wafer-Scale Cluster", "Condor Galaxy supercomputers" ], "note": "Cerebras Inference is the hosted API on the same hardware and is catalogued separately." } ], "versionsUpdated": "2026-07-29" }, { "id": "dx-h1", "name": "DX-H1", "company": "DEEPX", "region": "korea", "tagline": "DEEPX high-performance edge AI accelerator", "description": "DX-H1 is DEEPX's higher-performance inference accelerator for edge servers and industrial vision, sitting above the DX-M on-device line. DEEPX targets Korean and global OEM design wins in security, robotics and industrial inspection; shipment volumes and named production programmes are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://deepx.ai/products/dx-h1-v-npu/", "apiLink": null, "aiCoreShare": 86, "sourceUrls": [ "https://deepx.ai/products/dx-h1-v-npu/" ] }, { "id": "dojo", "name": "Dojo", "company": "Tesla", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.tesla.com/AI", "apiLink": null, "aiCoreShare": 80, "sourceUrls": [ "https://www.tesla.com/AI" ], "contentReview": "needs-review" }, { "id": "waymo-via", "name": "Waymo Via", "company": "Waymo", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://waymo.com", "apiLink": null, "aiCoreShare": 86, "contentReview": "needs-review" }, { "id": "huashan-a1000", "name": "Black Sesame Huashan", "company": "Black Sesame Technologies", "region": "china", "tagline": "Automotive AI compute platform for driver assistance", "description": "Huashan is Black Sesame Technologies' automotive-grade AI compute platform — the A1000 domain-controller silicon plus the autonomous-driving software stack built on it. Black Sesame is HKEX-listed and competes with Horizon Journey and NVIDIA Orin for Chinese OEM design wins. Shipment volumes and named production programmes are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "Waymo", "Tesla FSD", "Mobileye", "Cruise" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Black Sesame Autonomous Driving Platform" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://bst.ai", "apiLink": null, "aiCoreShare": 83, "sourceUrls": [], "versions": [ { "name": "Black Sesame Huashan", "variants": [ "Black Sesame Autonomous Driving Platform" ] } ], "scoringRationale": "78 reflects HKEX-listed automotive AI silicon competing for Chinese OEM design wins against Horizon Journey and NVIDIA Orin. Held at the cap because shipment volumes and named production programmes are not disclosed — the same standard applied to Korean automotive-silicon rows.", "versionsUpdated": "2026-07-29" }, { "id": "figure", "name": "Figure", "company": "Figure AI", "region": "global", "tagline": "Figure AI's general-purpose humanoid robot line", "description": "Figure is Figure AI's general-purpose humanoid robot line, developed through numbered generations and targeted at logistics, manufacturing and eventually home work. Generations are steps in one product line rather than separate products. Commercial deployment beyond pilot programs is not publicly quantified.", "users": "Pilot deployments", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "atlas", "tesla-optimus", "unitree-h1", "neo" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Figure 02", "168cm height, 70kg weight", "16 DOF dexterous hands", "5 hours operation", "Natural language interaction" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://figure.ai", "apiLink": null, "aiCoreShare": 84, "versions": [ { "name": "Figure 02", "released": "2024-08", "variants": [ "Figure 01", "Figure 02" ], "note": "Numbered generations of one humanoid line; deployment counts are not disclosed." } ], "sourceUrls": [], "versionsUpdated": "2026-07-29" }, { "id": "helix-vision-language-action-model", "name": "Helix Vision-Language-Action Model", "company": "Figure AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.figure.ai/helix", "apiLink": null, "aiCoreShare": 88, "sourceUrls": [ "https://www.figure.ai/helix" ], "contentReview": "needs-review" }, { "id": "tesla-optimus", "name": "Tesla Optimus", "company": "Tesla", "region": "global", "tagline": "Tesla's humanoid robot", "description": "Optimus is Tesla's general-purpose humanoid robot designed for manufacturing and eventually household tasks.", "users": "In development", "userCount": 0, "releaseDate": "2024-10", "technicalImpact": 80, "competitors": [ "figure", "atlas", "unitree-h1", "neo", "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "General-purpose humanoid", "Tesla AI integration", "Manufacturing tasks", "FSD hardware", "Suite modules: Optimus", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "target": "Target $20K-30K" }, "link": "https://tesla.com/optimus", "apiLink": null, "aiCoreShare": 80, "sourceUrls": [] }, { "id": "hyundai-waymo-ioniq-5-robotaxi-program", "name": "Hyundai-Waymo IONIQ 5 Robotaxi Program", "company": "Hyundai", "region": "korea", "tagline": "IONIQ 5 vehicle-platform integration for Waymo One autonomous ride-hailing.", "description": "Hyundai-Waymo IONIQ 5 Robotaxi Program is Hyundai's official vehicle-platform partnership with Waymo. It is not a standalone Hyundai autonomous-driving AI product; Waymo's sixth-generation autonomous driving technology is integrated into Hyundai IONIQ 5 vehicles for the Waymo One fleet.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "Tesla FSD", "Mobileye Drive", "Zoox", "Cruise" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.hyundaimotorgroup.com/amp/CONT0000000000161922", "apiLink": null, "aiCoreShare": 65, "adoptionSignal": "Official Hyundai-Waymo partnership; Hyundai product-specific robotaxi user count is not disclosed.", "sourceUrls": [ "https://www.hyundaimotorgroup.com/amp/CONT0000000000161922" ] }, { "id": "hyundai-mobed", "name": "MobED", "company": "Hyundai", "region": "korea", "tagline": "Hyundai Motor Group's mobile eccentric droid robotics platform.", "description": "MobED is Hyundai Motor Group Robotics LAB's production mobile robot platform. Hyundai describes MobED Pro as using AI-based algorithms plus LiDAR and camera sensor fusion for autonomous driving capability.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 77, "competitors": [ "Boston Dynamics Spot", "MiR AMR", "Agility Digit", "unitree-h1" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.hyundaimotorgroup.com/en/story/CONT0000000000199523", "apiLink": null, "aiCoreShare": 77, "sourceUrls": [ "https://www.hyundaimotorgroup.com/en/story/CONT0000000000199523", "https://www.hyundaimotorgroup.com/news/CONT0000000000198949" ], "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 82 -> 77. MobED Pro is a listed commercial product from a credible robotics group, but deployment volume, revenue and external customers are all undisclosed — 82 was the tier of Lunit INSIGHT (10,000+ sites) and ATOM-Max (CSAP-certified KT Cloud deployment). Unquantified commercial availability sits at the 77 tier." }, { "id": "unitree-h1", "name": "Unitree Humanoid", "company": "Unitree Robotics", "region": "china", "tagline": "Unitree's humanoid robot line", "description": "Unitree's humanoid family spans H2 Plus, H2, R1, G1, and H1 platforms sold to research, education, development, and industrial buyers. Unitree became a public STAR Market company in August 2026; its broad price range remains the family's primary market differentiator.", "users": "1K+", "userCount": 1000, "releaseDate": "2023-08", "technicalImpact": 84, "competitors": [ "figure", "atlas", "tesla-optimus", "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Full humanoid form", "Bipedal walking and running", "Up to 3.3 m/s speed", "SDK for development", "Suite modules: Unitree G1, Unitree R1", "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment" ], "pricing": { "commercial": "~$90,000" }, "link": "https://www.unitree.com/h1", "apiLink": "https://support.unitree.com", "aiCoreShare": 90, "scoringRationale": "Unitree's full-size humanoid robot sold to researchers, universities, and enterprise customers at $90K — dramatically below Boston Dynamics or Figure pricing. 1,000+ units shipped is a real, sourced deployment number. Score reflects breakthrough price-performance positioning and growing developer ecosystem. Kept at 84 because production-scale industrial deployments, revenue, and durable customer lock-in are not yet demonstrated.", "sourceUrls": [ "https://www.unitree.com/h1/", "https://techcrunch.com/2024/05/23/unitree-robotics-h1-humanoid-robot/", "https://shop.unitree.com/", "https://www.sse.com.cn/disclosure/announcement/listing/ipo/c/c_20260818_10829204.shtml" ], "versions": [ { "name": "Unitree H2 Plus", "released": "2026-06", "variants": [ "Unitree H2", "Unitree R1", "Unitree G1", "Unitree H1" ] }, { "name": "Unitree H1", "released": "2023-08" } ], "versionsUpdated": "2026-09-02" }, { "id": "waymo-driver", "name": "Waymo Driver", "company": "Waymo", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://waymo.com", "apiLink": null, "aiCoreShare": 87, "contentReview": "needs-review" }, { "id": "zoox-robotaxi", "name": "Zoox Robotaxi", "company": "Zoox", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "Waymo", "Tesla FSD", "Mobileye", "Cruise" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://zoox.com", "apiLink": null, "aiCoreShare": 80, "contentReview": "needs-review" }, { "id": "sambanova-cloud", "name": "SambaNova Cloud", "company": "SambaNova", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://sambanova.ai", "apiLink": null, "aiCoreShare": 79, "contentReview": "needs-review" }, { "id": "waymo-one", "name": "Waymo One", "company": "Waymo", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://waymo.com", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "geekplus-sorting-robots", "name": "Geek+ Warehouse Robotics", "company": "Geekplus", "region": "china", "tagline": "Autonomous mobile robots and orchestration for warehouse fulfilment", "description": "Geek+ builds warehouse automation — goods-to-person mobile robots, sorting and picking systems, and the shuttle and orchestration software that schedules them across a facility. It is HKEX-listed and one of the largest AMR vendors globally by deployment, competing with Symbotic and GreyOrange. Per-product installed fleet sizes and named customer counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 77, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Geekplus Picking Robots, Geekplus Warehouse AI, Geekplus RoboShuttle" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://geekplus.com", "apiLink": null, "aiCoreShare": 90, "sourceUrls": [], "versions": [ { "name": "Geek+ Warehouse Robotics", "variants": [ "Geekplus Picking Robots", "Geekplus Warehouse AI", "Geekplus RoboShuttle" ] } ], "scoringRationale": "77 reflects HKEX-listed status and one of the largest global AMR deployment footprints, competing with Symbotic and GreyOrange. Held below 80 because installed fleet sizes, named customers and revenue attribution are not recorded in the catalogue.", "versionsUpdated": "2026-07-29" }, { "id": "sambanova-datascale", "name": "SambaNova DataScale", "company": "SambaNova", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://sambanova.ai", "apiLink": null, "aiCoreShare": 78, "contentReview": "needs-review" }, { "id": "greyorange-gstore", "name": "GreyOrange gStore", "company": "GreyOrange", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://greyorange.com", "apiLink": null, "aiCoreShare": 85, "contentReview": "needs-review" }, { "id": "flowstate", "name": "Flowstate", "company": "Intrinsic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.intrinsic.ai/flowstate", "apiLink": null, "aiCoreShare": 86, "sourceUrls": [ "https://www.intrinsic.ai/flowstate" ], "contentReview": "needs-review" }, { "id": "symbotic-system", "name": "Symbotic System", "company": "Symbotic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://symbotic.com", "apiLink": null, "aiCoreShare": 86, "contentReview": "needs-review" }, { "id": "eve", "name": "EVE", "company": "1X", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.1x.tech/eve", "apiLink": null, "aiCoreShare": 82, "sourceUrls": [ "https://www.1x.tech/eve" ], "contentReview": "needs-review" }, { "id": "atlas", "name": "Atlas", "company": "Boston Dynamics", "region": "global", "tagline": "World's most dynamic humanoid", "description": "Atlas is Boston Dynamics' research humanoid robot showcasing advanced mobility with parkour and complex manipulation.", "users": "Research partners", "userCount": 0, "releaseDate": "2013-07", "technicalImpact": 80, "competitors": [ "figure", "tesla-optimus", "unitree-h1" ], "features": [ "Full-body mobility", "Electric actuation", "Advanced balance", "Parkour and acrobatics" ], "pricing": { "research": "Contact Boston Dynamics" }, "link": "https://bostondynamics.com/atlas", "apiLink": null, "aiCoreShare": 76 }, { "id": "deepx-sdk", "name": "DEEPX SDK", "company": "DEEPX", "region": "korea", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://deepx.ai", "apiLink": null, "aiCoreShare": 84, "scoringRationale": "76 reflects the toolchain that determines whether DEEPX silicon is practically adoptable — compiler, runtime and framework support. It is scored as developer infrastructure rather than under the hardware cap, and held at the default tier because no developer or design-win figures are disclosed.", "contentReview": "needs-review" }, { "id": "doosan-robot-lab-ai", "name": "Doosan Robotics AI", "company": "Doosan Robotics", "region": "korea", "tagline": "AI software and programming stack for Doosan collaborative robots", "description": "Doosan Robotics pairs its collaborative robot arms with the Dart-Suite programming environment and AI capabilities for vision-guided handling and task learning, so integrators can deploy cobots without conventional robot programming. Doosan Robotics is KOSPI-listed; AI-attributable revenue and named deployments are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment", "Suite modules: Doosan Cobots AI, Doosan Dart-Suite" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://doosanrobotics.com", "apiLink": null, "aiCoreShare": 83, "sourceUrls": [], "versions": [ { "name": "Doosan Robotics AI", "variants": [ "Doosan Cobots AI", "Doosan Dart-Suite" ] } ], "scoringRationale": "76 reflects KOSPI-listed Doosan Robotics pairing cobot hardware with a programming environment that removes conventional robot programming from deployment. Held at the default tier because AI-attributable revenue and named deployments are not disclosed.", "versionsUpdated": "2026-07-29" }, { "id": "greymatter", "name": "GreyMatter", "company": "GreyOrange", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.greyorange.com/greymatter/", "apiLink": null, "aiCoreShare": 81, "sourceUrls": [ "https://www.greyorange.com/greymatter/" ], "contentReview": "needs-review" }, { "id": "rainbow-robotics-humanoid-platform", "name": "Rainbow Robotics", "company": "Rainbow Robotics", "region": "korea", "tagline": "Korean humanoid and manipulation robotics platform", "description": "Rainbow Robotics builds humanoid and collaborative robot platforms with the manipulation and control software that runs them, spun out of KAIST and now majority-linked to Samsung Electronics, which gives it an unusual industrial path to deployment for a Korean robotics firm. Unit shipments and named production deployments are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "figure", "1X NEO", "Tesla Optimus", "Boston Dynamics Atlas" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment", "Suite modules: Rainbow Robotics Manipulation AI" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://rainbowrobotics.com", "apiLink": null, "aiCoreShare": 80, "sourceUrls": [], "versions": [ { "name": "Rainbow Robotics", "variants": [ "Rainbow Robotics Manipulation AI" ] } ], "scoringRationale": "76 reflects a KAIST-origin humanoid platform with an unusual industrial path through Samsung Electronics' majority stake — real distribution potential for a Korean robotics firm. Held at the default tier because unit shipments and named production deployments are not disclosed.", "versionsUpdated": "2026-07-29" }, { "id": "1x-world-model", "name": "1X World Model", "company": "1X", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://1x.tech", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "dx-m", "name": "DX-M NPU M1/M2", "company": "DEEPX", "region": "korea", "tagline": "DEEPX's DX-M on-device inference NPU series — DX-M1 and DX-M2", "description": "DX-M is DEEPX's on-device inference NPU series for edge vision and embedded AI, spanning the DX-M1 and the higher-capacity DX-M2. The series is what an OEM designs in, so the parts share one row. DEEPX does not disclose shipment volumes or named production programs for either part.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: DX-M2" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://deepx.ai/products/dx-m1/", "apiLink": null, "aiCoreShare": 83, "sourceUrls": [ "https://deepx.ai/products/dx-m1/", "https://deepx.ai/product/dx-m2/" ], "versions": [ { "name": "DX-M series", "variants": [ "DX-M1", "DX-M2" ], "note": "On-device inference NPUs for edge vision and embedded AI." } ], "versionsUpdated": "2026-07-29" }, { "id": "1x-teleoperation-platform", "name": "1X Teleoperation Platform", "company": "1X", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://1x.tech", "apiLink": null, "aiCoreShare": 89, "contentReview": "needs-review" }, { "id": "d-matrix-inference-platform", "name": "d-Matrix Inference Platform", "company": "d-Matrix", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Suite modules: d-Matrix DIMC" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://d-matrix.ai", "apiLink": null, "aiCoreShare": 88, "sourceUrls": [], "versions": [ { "name": "d-Matrix Inference Platform", "variants": [ "d-Matrix DIMC" ] } ], "versionsUpdated": "2026-07-29", "contentReview": "needs-review" }, { "id": "rldx-1", "name": "RLDX-1", "company": "RLWRLD", "region": "korea", "tagline": "Dexterity-first foundation model for robot hands", "description": "RLDX-1 is RLWRLD's dexterity-first foundation model for robot hands. The model targets real-world industrial manipulation with high-DoF hands, combining vision-language reasoning, motion, memory, tactile, and torque signals for dexterous robotics tasks.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2026-05-07", "technicalImpact": 76, "competitors": [ "nvidia-isaac", "covariant", "figure-ai", "covariant-brain", "figure-humanoid-workforce", "Physical Intelligence" ], "features": [ "Dexterous robot-hand policy model", "Vision-language action architecture", "Motion, memory, tactile, and torque streams", "Real-time robotics inference optimization" ], "pricing": { "enterprise": "Contact RLWRLD" }, "link": "https://www.rlwrld.ai/ko/rldx-1", "apiLink": null, "aiCoreShare": 95, "adoptionSignal": "RLWRLD publishes RLDX-1 as a research/model release with a technical report, code, and Hugging Face checkpoints; product-specific users, customers, or deployments are not disclosed.", "sourceUrls": [ "https://www.rlwrld.ai/ko/rldx-1", "https://arxiv.org/abs/2605.03269", "https://github.com/RLWRLD/RLDX-1", "https://huggingface.co/collections/RLWRLD/rldx-1" ], "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 80 -> 76. Promising dexterous-manipulation research with open checkpoints, but no customers, deployments or usage — pre-adoption rows sit at the default tier regardless of strategic relevance." }, { "id": "orbit", "name": "Orbit", "company": "Boston Dynamics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://bostondynamics.com/products/orbit/", "apiLink": null, "aiCoreShare": 85, "sourceUrls": [ "https://bostondynamics.com/products/orbit/" ], "contentReview": "needs-review" }, { "id": "arion-smet", "name": "Arion-SMET", "company": "Hanwha", "region": "korea", "tagline": "Hanwha Aerospace multipurpose unmanned ground vehicle.", "description": "Arion-SMET is Hanwha Aerospace's proprietary multipurpose unmanned ground vehicle for autonomous and robotic off-road logistics and support missions.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.hanwha.com/companies/hanwha-aerospace.do?nlsFlag=gnb", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://www.hanwha.com/companies/hanwha-aerospace.do?nlsFlag=gnb", "https://www.hanwha.com/newsroom/news/press-releases/hanwha-showcases-unmanned-and-ai-capabilities-at-bsda-2026-to-advance-european-partnerships.do" ] }, { "id": "walker-x", "name": "UBTech Walker", "company": "UBTech", "region": "china", "tagline": "UBTech's humanoid robot line for service and industrial work", "description": "Walker is UBTech's humanoid robot line, spanning service and demonstration deployments and the industrial variants pitched at Chinese manufacturing pilots, covering bipedal locomotion, manipulation and voice interaction. UBTech is HKEX-listed and has announced factory pilots with Chinese manufacturers; unit deliveries and paying-customer counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5", "figure", "1X NEO", "Tesla Optimus", "Boston Dynamics Atlas" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Walker S, UBTech Industrial Humanoid" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://ubtrobot.com", "apiLink": null, "aiCoreShare": 91, "sourceUrls": [], "versions": [ { "name": "UBTech Walker", "variants": [ "Walker S", "UBTech Industrial Humanoid" ] } ], "scoringRationale": "79 reflects a full-size humanoid line from an HKEX-listed manufacturer with announced industrial pilots at Chinese factories, which is more than a research platform. Held below 84 because UBTech discloses no unit deliveries, paying-customer counts or pilot conversion, and humanoid deployment claims across the industry outrun verifiable installations.", "versionsUpdated": "2026-07-29" }, { "id": "unitree-go2", "name": "Unitree Go2", "company": "Unitree Robotics", "region": "china", "tagline": "Affordable quadruped robot", "description": "Go2 is Unitree's consumer and professional quadruped offering Boston Dynamics-like capabilities at a fraction of the price.", "users": "50K+", "userCount": 50000, "releaseDate": "2023-07", "technicalImpact": 79, "competitors": [ "spot", "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct" ], "features": [ "Advanced locomotion", "GPT integration", "4D LiDAR", "Up to 3.5 m/s speed" ], "pricing": { "air": "$1,600", "pro": "$2,800", "edu": "$3,500" }, "link": "https://www.unitree.com/go2", "apiLink": "https://support.unitree.com", "aiCoreShare": 78, "scoringRationale": "Unitree's AI-enabled quadruped robot at $1,600–$3,500 — the most accessible legged robot for education, research, and light commercial use. 50,000+ units shipped. Score reflects strong price-performance, real deployment scale, and developer community. Kept below 84 because use cases remain research/hobbyist-oriented and enterprise production deployments are limited.", "sourceUrls": [ "https://www.unitree.com/go2/" ] }, { "id": "stretch", "name": "Stretch", "company": "Boston Dynamics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://bostondynamics.com/products/stretch/", "apiLink": null, "aiCoreShare": 81, "sourceUrls": [ "https://bostondynamics.com/products/stretch/" ], "contentReview": "needs-review" }, { "id": "xeon-ai-acceleration", "name": "Xeon AI Acceleration", "company": "Intel", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.intel.com/content/www/us/en/products/details/processors/xeon.html", "apiLink": null, "aiCoreShare": 85, "sourceUrls": [ "https://www.intel.com/content/www/us/en/products/details/processors/xeon.html" ], "contentReview": "needs-review" }, { "id": "skild-foundation-model-for-robotics", "name": "Skild Foundation Model for Robotics", "company": "Skild AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://skild.ai", "apiLink": null, "aiCoreShare": 79, "contentReview": "needs-review" }, { "id": "aurora-connect", "name": "Aurora Connect", "company": "Aurora", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 77, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://aurora.tech", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "cambricon-cloud-ai-chip", "name": "Cambricon MLU", "company": "Cambricon", "region": "china", "tagline": "Cambricon's datacentre AI accelerators and Neuware toolchain", "description": "Cambricon designs MLU datacentre AI accelerators and the Neuware software stack — compilers, operator libraries and framework plugins — that lets PyTorch and TensorFlow workloads target them. It is STAR-listed and one of the main domestic alternatives to NVIDIA in Chinese datacentres. Shipment volumes and named deployments are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Edge or embodied AI", "Inference acceleration", "Autonomous operation", "Industrial deployment", "Suite modules: Cambricon Neuware" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://cambricon.com", "apiLink": null, "aiCoreShare": 90, "sourceUrls": [], "versions": [ { "name": "Cambricon MLU", "variants": [ "Cambricon Neuware" ] } ], "scoringRationale": "78 applies the tiered AI-hardware cap at tier 0: Cambricon is STAR-listed with a real datacentre accelerator line and its own Neuware toolchain, but the catalogue records no named cloud or OEM deployment and no shipment volume. Clears to 82 with one named deployment, 86 with multi-partner reach or disclosed volume.", "versionsUpdated": "2026-07-29" }, { "id": "see-and-spray", "name": "See & Spray", "company": "John Deere", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 77, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.deere.com/en/sprayers/see-spray-gen-2/", "apiLink": null, "aiCoreShare": 87, "sourceUrls": [ "https://www.deere.com/en/sprayers/see-spray-gen-2/" ], "contentReview": "needs-review" }, { "id": "apptronik-humanoid-platform", "name": "Apptronik Humanoid Platform", "company": "Apptronik", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "figure", "1X NEO", "Tesla Optimus", "Boston Dynamics Atlas" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://apptronik.com", "apiLink": null, "aiCoreShare": 85, "contentReview": "needs-review" }, { "id": "unitree-b2", "name": "Unitree B2", "company": "Unitree Robotics", "region": "china", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://unitree.com", "apiLink": null, "aiCoreShare": 83, "scoringRationale": "76 reflects a quadruped platform with meaningful research and industrial-inspection adoption and aggressive pricing against Boston Dynamics Spot. Held at the default tier because Unitree discloses no unit shipments for the B-series.", "contentReview": "needs-review" }, { "id": "amd-instinct", "name": "Instinct GPU MI300X/MI350", "company": "AMD", "region": "global", "tagline": "AMD's Instinct datacenter AI accelerator line — MI300 and MI350 generations", "description": "Instinct is AMD's datacenter AI accelerator line and the main non-NVIDIA option for large-model training and inference, paired with the ROCm software stack. Buyers evaluate the line and its current generation rather than individual part numbers, so the MI300 and MI350 generations share one row. AMD does not disclose per-part shipment or customer counts.", "users": "Enterprise hyperscalers", "userCount": 0, "releaseDate": "2023-12-06", "technicalImpact": 83, "competitors": [ "h100-tensor-core-gpu", "nvidia-hopper", "NVIDIA Blackwell", "Cloud TPU v5", "Intel Gaudi 3" ], "features": [ "Datacenter AI training and inference accelerators", "High-bandwidth memory capacity per accelerator", "ROCm open software stack", "Cloud and OEM system availability", "Primary non-NVIDIA datacenter accelerator alternative", "Suite modules: Instinct MI350", "Large-model training", "AI inference acceleration", "Improved compute density vs MI300X", "Cloud and enterprise deployment" ], "pricing": { "free": "N/A", "pro": "N/A", "enterprise": "Contact AMD or cloud provider" }, "link": "https://www.amd.com/en/products/accelerators/instinct.html", "apiLink": null, "aiCoreShare": 90, "adoptionSignal": "Deployed in Microsoft Azure ND MI300X v5 instances, Meta AI production inference workloads, and Oracle Cloud Infrastructure; hyperscaler-scale deployments confirmed.", "scoringRationale": "Raised to 83 per infrastructure-backbone rule: cloud/OEM deployment evidence replaces app-style user counts for accelerator hardware. Azure, Meta, and Oracle deployments confirm production hyperscaler scale. AMD holds #2 position in AI accelerators with real switching leverage. Capped below H100 (93) reflecting CUDA ecosystem advantage, broader deployment breadth, and stronger developer tooling at NVIDIA.", "sourceUrls": [ "https://www.amd.com/en/products/accelerators/instinct/mi300.html", "https://azure.microsoft.com/en-us/blog/azure-unveils-new-ai-infrastructure-with-amd-mi300x-gpus/", "https://www.amd.com/en/products/accelerators/instinct/mi350.html" ], "versions": [ { "name": "Instinct MI350 series", "released": "2025", "variants": [ "MI350X", "MI355X" ], "note": "Current generation." }, { "name": "Instinct MI300 series", "released": "2023-12", "variants": [ "MI300X", "MI325X" ], "note": "Widely deployed prior generation." } ], "versionsUpdated": "2026-07-29" }, { "id": "carbon-autonomy-platform", "name": "Carbon Autonomy Platform", "company": "Carbon Robotics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 75, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://carbonrobotics.com", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "john-deere-operations-center-ai", "name": "John Deere Operations Center AI", "company": "John Deere", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 75, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://deere.com", "apiLink": null, "aiCoreShare": 78, "contentReview": "needs-review" }, { "id": "physical-intelligence-robot-foundation-model", "name": "Physical Intelligence Robot Foundation Model", "company": "Physical Intelligence", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 75, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://physicalintelligence.company", "apiLink": null, "aiCoreShare": 87, "contentReview": "needs-review" }, { "id": "symbotic-warehouse-ai", "name": "Symbotic Warehouse AI", "company": "Symbotic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 75, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://symbotic.com", "apiLink": null, "aiCoreShare": 82, "contentReview": "needs-review" }, { "id": "tenstorrent-ai-cloud", "name": "Tenstorrent AI Cloud", "company": "Tenstorrent", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 75, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://tenstorrent.com", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "openvino", "name": "OpenVINO", "company": "Intel", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 74, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/overview.html", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/overview.html" ], "contentReview": "needs-review" }, { "id": "skild-brain", "name": "Skild Brain", "company": "Skild AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 74, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://skild.ai", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "bone-autonomy-stack", "name": "Autonomy Stack", "company": "Bone AI", "region": "korea", "tagline": "End-to-end autonomy software for drones and vehicles", "description": "Bone AI's Autonomy Stack provides end-to-end autonomy software, command/control, and mission planning for autonomous systems (UAV, UGV, USV). Based in Seoul and Palo Alto. Traction in defense, logistics, wildfire detection, and anti-drone missions.", "users": "Government contracts", "userCount": 0, "releaseDate": "2025-01", "technicalImpact": 73, "competitors": [ "anduril", "helsing", "skydio" ], "features": [ "UAV/UGV/USV autonomy", "Mission planning", "Command and control", "Defense applications" ], "pricing": { "government": "B2G contracts" }, "link": "https://www.bonerobotics.ai", "apiLink": null, "aiCoreShare": 74, "scoringRationale": "73 reflects an autonomy software stack aimed at industrial and delivery robots. Held below the default tier because Bone AI discloses no deployments, customers or fleet figures, and autonomy stacks are judged on production proof rather than capability claims." }, { "id": "laserweeder", "name": "LaserWeeder", "company": "Carbon Robotics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 73, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://carbonrobotics.com/laserweeder/", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://carbonrobotics.com/laserweeder/" ], "contentReview": "needs-review" }, { "id": "wudang-c1200", "name": "Wudang C1200", "company": "Black Sesame Technologies", "region": "china", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://bst.ai", "apiLink": null, "aiCoreShare": 82, "scoringRationale": "72 reflects a second Black Sesame automotive compute line targeting cross-domain vehicle control. Held below the default tier because it has less disclosed design-win traction than Huashan and no shipment or programme evidence.", "contentReview": "needs-review" }, { "id": "autonomous-8r-tractor", "name": "Autonomous 8R Tractor", "company": "John Deere", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.deere.com/en/autonomous/", "apiLink": null, "aiCoreShare": 85, "sourceUrls": [ "https://www.deere.com/en/autonomous/" ], "contentReview": "needs-review" }, { "id": "physical-intelligence-manipulation-policy", "name": "Physical Intelligence Manipulation Policy", "company": "Physical Intelligence", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://physicalintelligence.company", "apiLink": null, "aiCoreShare": 88, "contentReview": "needs-review" }, { "id": "sanctuary-teleoperation", "name": "Sanctuary Teleoperation", "company": "Sanctuary AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://sanctuary.ai", "apiLink": null, "aiCoreShare": 79, "contentReview": "needs-review" }, { "id": "carbon-ops-center", "name": "Carbon Ops Center", "company": "Carbon Robotics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 71, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://carbonrobotics.com", "apiLink": null, "aiCoreShare": 83, "contentReview": "needs-review" }, { "id": "kia-highway-driving-assist-2", "name": "Highway Driving Assist", "company": "Kia", "region": "korea", "tagline": "Kia's highway driver-assistance feature set.", "description": "Highway Driving Assist 2 is Kia's driver-assistance feature for highway speed, distance, lane centering, and lane-change support where fitted.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 69, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://worldwide.kia.com/en/innovation/future-mobility", "apiLink": null, "aiCoreShare": 72, "sourceUrls": [ "https://worldwide.kia.com/en/innovation/future-mobility", "https://www.kia.com/ie/service/safety/adas/" ], "versions": [ { "name": "Highway Driving Assist 2" } ], "versionsUpdated": "2026-07-29" }, { "id": "eyeq-ultra", "name": "EyeQ Ultra", "company": "Mobileye", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 71, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.mobileye.com/technology/eyeq-chip/", "apiLink": null, "aiCoreShare": 82, "sourceUrls": [ "https://www.mobileye.com/technology/eyeq-chip/" ], "contentReview": "needs-review" }, { "id": "cruzr", "name": "Cruzr", "company": "UBTech", "region": "china", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 71, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://ubtrobot.com", "apiLink": null, "aiCoreShare": 77, "scoringRationale": "71 reflects a commercial service robot with retail and public-space deployments, distinct from UBTech's humanoid line. Held below the default tier because unit deliveries and customer counts are not disclosed.", "contentReview": "needs-review" }, { "id": "spot", "name": "Spot", "company": "Boston Dynamics", "region": "global", "tagline": "The agile mobile robot", "description": "Spot is Boston Dynamics' quadruped robot for industrial inspection, data collection, and hazardous environment operation.", "users": "1500+ deployed", "userCount": 1500, "releaseDate": "2020-06", "technicalImpact": 70, "competitors": [ "unitree-go2", "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct" ], "features": [ "360° perception", "Stair climbing", "Autonomous navigation", "14kg payload" ], "pricing": { "purchase": "$74,500+ base", "lease": "Leasing available" }, "link": "https://bostondynamics.com/products/spot", "apiLink": "https://dev.bostondynamics.com", "aiCoreShare": 76 }, { "id": "corsair", "name": "Corsair", "company": "d-Matrix", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 70, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.d-matrix.ai/corsair-at-rack-scale/", "apiLink": null, "aiCoreShare": 84, "sourceUrls": [ "https://www.d-matrix.ai/corsair-at-rack-scale/" ], "contentReview": "needs-review" }, { "id": "agility-arc", "name": "Agility Arc", "company": "Agility Robotics", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 69, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://agilityrobotics.com", "apiLink": null, "aiCoreShare": 82, "contentReview": "needs-review" }, { "id": "skypod", "name": "Skypod", "company": "Exotec", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 68, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.exotec.com/en/solutions/skypod-system/", "apiLink": null, "aiCoreShare": 90, "sourceUrls": [ "https://www.exotec.com/en/solutions/skypod-system/" ], "contentReview": "needs-review" }, { "id": "gaudi-3", "name": "Gaudi", "company": "Intel", "region": "global", "tagline": "Intel's Gaudi AI training and inference accelerator series", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html", "apiLink": null, "aiCoreShare": 90, "sourceUrls": [ "https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html" ], "versions": [ { "name": "Gaudi 3", "released": "2024", "note": "Current generation." }, { "name": "Gaudi 2", "released": "2022" } ], "versionsUpdated": "2026-07-29", "contentReview": "needs-review" }, { "id": "kia-pbv-software-platform", "name": "Kia PBV Software Platform", "company": "Kia", "region": "korea", "tagline": "Software and connected-services platform for Kia PBV vehicles.", "description": "Kia PBV Software Platform covers Kia's software-defined PBV experience, including PBV infotainment, app-store, OTA, fleet/data services, and AI voice-recognition capabilities.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://worldwide.kia.com/en/innovation/software-and-ai", "apiLink": null, "aiCoreShare": 70, "sourceUrls": [ "https://worldwide.kia.com/en/innovation/software-and-ai", "https://www.kia.com/dm/discover-kia/pbv/pbv-about.html" ] }, { "id": "samba-1", "name": "Samba-1", "company": "SambaNova", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://sambanova.ai", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "symbotic-breakpack", "name": "Symbotic BreakPack", "company": "Symbotic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://symbotic.com", "apiLink": null, "aiCoreShare": 79, "contentReview": "needs-review" }, { "id": "mobileye-chauffeur", "name": "Mobileye Chauffeur", "company": "Mobileye", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 66, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://mobileye.com", "apiLink": null, "aiCoreShare": 78, "contentReview": "needs-review" }, { "id": "zoox-autonomy-stack", "name": "Zoox Autonomy Stack", "company": "Zoox", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 66, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://zoox.com", "apiLink": null, "aiCoreShare": 85, "contentReview": "needs-review" }, { "id": "aurora-driver", "name": "Aurora Driver", "company": "Aurora", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 65, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://aurora.tech", "apiLink": null, "aiCoreShare": 91, "contentReview": "needs-review" }, { "id": "etched-transformer-asic", "name": "Etched Transformer ASIC", "company": "Etched", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 65, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://etched.com", "apiLink": null, "aiCoreShare": 86, "contentReview": "needs-review" }, { "id": "hanwha-hcr-cobots", "name": "Hanwha HCR Collaborative Robots", "company": "Hanwha", "region": "korea", "tagline": "Hanwha Robotics collaborative robot line.", "description": "Hanwha HCR Collaborative Robots are Hanwha Robotics' cobot products for industrial and service automation. Hanwha says the HCR line is combined with smart technologies including AI and sensors.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 65, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [ "Embodied AI", "Edge inference", "Autonomous operation", "Industrial deployment" ], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.hanwha.com/companies/hanwha-robotics.do", "apiLink": null, "aiCoreShare": 76, "sourceUrls": [ "https://www.hanwha.com/companies/hanwha-robotics.do" ] }, { "id": "ranger-assist", "name": "Ranger Assist", "company": "GreyOrange", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 64, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.greyorange.com/greymatter/assisted-picking/", "apiLink": null, "aiCoreShare": 87, "sourceUrls": [ "https://www.greyorange.com/greymatter/assisted-picking/" ], "contentReview": "needs-review" }, { "id": "intrinsic-skills", "name": "Intrinsic Skills", "company": "Intrinsic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 64, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://intrinsic.ai", "apiLink": null, "aiCoreShare": 78, "contentReview": "needs-review" }, { "id": "mujin-truckbot", "name": "Mujin TruckBot", "company": "Mujin", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 64, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://mujin-corp.com", "apiLink": null, "aiCoreShare": 89, "contentReview": "needs-review" }, { "id": "neo", "name": "NEO", "company": "1X Technologies", "region": "global", "tagline": "Safe humanoid for human environments", "description": "NEO is 1X Technologies' humanoid designed for safe operation alongside humans, backed by OpenAI and Samsung.", "users": "Pilot deployments", "userCount": 0, "releaseDate": "2024-01", "technicalImpact": 62, "competitors": [ "figure", "tesla-optimus", "atlas", "unitree-h1" ], "features": [ "Human-safe design", "Compliant actuation", "Home navigation", "OpenAI integration" ], "pricing": { "commercial": "Contact 1X" }, "link": "https://www.1x.tech/neo", "apiLink": null, "aiCoreShare": 76, "sourceUrls": [ "https://www.1x.tech/neo" ] }, { "id": "ryzen-ai", "name": "Ryzen AI", "company": "AMD", "region": "global", "tagline": "AMD's integrated NPU for consumer and commercial laptops, enabling Copilot+ PC and on-device AI features.", "description": "Ryzen AI is AMD's neural processing unit (NPU) integrated into consumer and commercial Ryzen laptop processors. Powers on-device AI workloads including Windows Copilot+ features, local LLM inference via AMD's Ryzen AI Software SDK, and video/creative task acceleration. Ships in hundreds of laptop models from Lenovo, HP, Dell, and ASUS. Competes with Qualcomm Snapdragon X Elite and Intel Core Ultra for the Copilot+ AI PC tier; benchmark leadership shifts between generations. On-device AI SDK and ISV ecosystem still maturing relative to Qualcomm's developer ecosystem.", "users": "Consumer and commercial laptop buyers", "userCount": 0, "releaseDate": "2023-06-01", "technicalImpact": 62, "competitors": [ "Qualcomm Snapdragon X Elite", "Intel Core Ultra", "Apple Neural Engine", "MediaTek Dimensity AI" ], "features": [ "On-device LLM inference", "Copilot+ PC compatibility", "NPU acceleration", "OEM laptop deployment" ], "pricing": { "free": "N/A", "pro": "N/A", "enterprise": "OEM integrated" }, "link": "https://www.amd.com/en/products/processors/consumer/ryzen-ai.html", "apiLink": null, "aiCoreShare": 83, "adoptionSignal": "Ships in consumer and commercial laptops from Lenovo, HP, Dell, and ASUS; Copilot+ PC tier compatibility across Ryzen AI 300 series.", "scoringRationale": "Scored 62: competitive on-device NPU with broad OEM distribution but constrained by a maturing SDK/ISV ecosystem vs. Qualcomm Snapdragon X Elite. AI PC market is early; adoption driven by OEM bundling rather than active developer pull.", "sourceUrls": [ "https://www.amd.com/en/products/processors/consumer/ryzen-ai.html" ] }, { "id": "zoox-fleet-operations", "name": "Zoox Fleet Operations", "company": "Zoox", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 62, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://zoox.com", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "rocm", "name": "ROCm", "company": "AMD", "region": "global", "tagline": "AMD's open-source GPU computing platform — the software layer enabling AI and HPC workloads on Instinct accelerators.", "description": "ROCm (Radeon Open Compute) is AMD's open-source software platform for GPU-accelerated AI and HPC workloads. Provides HIP (a CUDA-compatible programming model), AI/HPC libraries (MIOpen, rocBLAS, rocSPARSE), and runtime support for Instinct MI300X/MI350 accelerators. The enabling software layer for PyTorch, TensorFlow, and JAX workloads on AMD hardware. Ecosystem depth is the primary competitive gap versus CUDA: most AI frameworks and operators default to CUDA paths, and ROCm compatibility gaps surface for niche operations and third-party libraries. Active investment in HIP porting compatibility and PyTorch first-class support.", "users": "AI researchers and enterprise GPU users", "userCount": 0, "releaseDate": "2016-01-01", "technicalImpact": 61, "competitors": [ "CUDA", "Intel oneAPI", "OpenCL", "Apple Metal" ], "features": [ "CUDA-compatible HIP programming model", "PyTorch and TensorFlow support", "Open-source GPU runtime", "HPC and AI library stack" ], "pricing": { "free": "Open source", "pro": "N/A", "enterprise": "N/A" }, "link": "https://rocm.docs.amd.com/en/latest/", "apiLink": null, "aiCoreShare": 77, "adoptionSignal": "Official support in PyTorch and TensorFlow; deployed as the enabling software layer in hyperscaler AMD MI300X instances. Open-source contributor base active on GitHub.", "scoringRationale": "Scored 61: essential enabling software for AMD AI hardware but ecosystem maturity significantly trails CUDA. Most AI developers default to CUDA paths; ROCm compatibility gaps create real deployment friction. Score reflects structural importance as the software moat for AMD accelerators without overstating ecosystem depth.", "sourceUrls": [ "https://rocm.docs.amd.com/en/latest/" ] }, { "id": "intrinsic-robotics-platform", "name": "Intrinsic Robotics Platform", "company": "Intrinsic", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 61, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://intrinsic.ai", "apiLink": null, "aiCoreShare": 83, "contentReview": "needs-review" }, { "id": "mobileye-drive", "name": "Mobileye Drive", "company": "Mobileye", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 61, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://mobileye.com", "apiLink": null, "aiCoreShare": 76, "contentReview": "needs-review" }, { "id": "mujin-robot-intelligence", "name": "Mujin Robot Intelligence", "company": "Mujin", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 61, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://mujin-corp.com", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "apollo", "name": "Apollo", "company": "Apptronik", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 60, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://apptronik.com/apollo", "apiLink": null, "aiCoreShare": 89, "sourceUrls": [ "https://apptronik.com/apollo" ], "contentReview": "needs-review" }, { "id": "mobileye-supervision", "name": "Mobileye SuperVision", "company": "Mobileye", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 60, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://mobileye.com", "apiLink": null, "aiCoreShare": 84, "contentReview": "needs-review" }, { "id": "core-ultra-npu", "name": "Core Ultra NPU", "company": "Intel", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 59, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://www.intel.com/content/www/us/en/products/details/processors/core-ultra.html", "apiLink": null, "aiCoreShare": 85, "sourceUrls": [ "https://www.intel.com/content/www/us/en/products/details/processors/core-ultra.html" ], "contentReview": "needs-review" }, { "id": "skild-simulation-platform", "name": "Skild Simulation Platform", "company": "Skild AI", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 59, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://skild.ai", "apiLink": null, "aiCoreShare": 87, "contentReview": "needs-review" }, { "id": "aurora-horizon", "name": "Aurora Horizon", "company": "Aurora", "region": "global", "tagline": "", "description": "", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 57, "competitors": [ "NVIDIA Hopper", "NVIDIA Blackwell", "AMD Instinct", "Cloud TPU v5" ], "features": [], "pricing": { "free": "Varies by plan", "pro": "Subscription or usage-based", "enterprise": "Contact sales" }, "link": "https://aurora.tech", "apiLink": null, "aiCoreShare": 81, "contentReview": "needs-review" }, { "id": "versal-ai-edge", "name": "Versal AI Edge", "company": "AMD", "region": "global", "tagline": "AMD's adaptive SoC/FPGA platform for edge AI inference in industrial, automotive, and aerospace deployments.", "description": "Versal AI Edge is AMD's adaptive SoC series combining FPGA fabric, ARM Cortex processors, and dedicated AI Engines (VLIW/SIMD vector processors) for edge inference workloads. Targets latency-sensitive, power-constrained deployments in industrial automation, autonomous vehicles, aerospace, and medical imaging. Programmable FPGA fabric enables hardware-level customization for specific AI inference pipelines — a differentiator versus fixed GPU edge accelerators. Competes with NVIDIA Jetson in embedded AI but targets higher-complexity, lower-volume industrial applications where hardware reconfigurability matters more than raw throughput.", "users": "Industrial and embedded AI developers", "userCount": 0, "releaseDate": "2022-01-01", "technicalImpact": 56, "competitors": [ "nvidia-jetson-orin", "nvidia-jetson-thor", "Intel Agilex", "Hailo-8", "Mobileye EyeQ" ], "features": [ "FPGA-accelerated edge inference", "AI Engines for DSP and ML workloads", "Industrial and automotive deployment", "Hardware reconfigurability" ], "pricing": { "free": "N/A", "pro": "N/A", "enterprise": "Contact AMD" }, "link": "https://www.amd.com/en/products/adaptive-socs-and-fpgas/versal/ai-edge-series.html", "apiLink": null, "aiCoreShare": 90, "adoptionSignal": "Production deployments in industrial automation, aerospace, and automotive OEM applications; specific customer names not publicly disclosed.", "scoringRationale": "Scored 56: niche edge AI FPGA/SoC platform with real industrial deployments but limited volume compared to GPU-based edge inference. FPGA reconfigurability is a genuine differentiator for specific verticals but constrains addressable market breadth.", "sourceUrls": [ "https://www.amd.com/en/products/adaptive-socs-and-fpgas/versal/ai-edge-series.html" ] }, { "id": "42dot-umos", "name": "UMOS", "nativeName": "유모스", "company": "42dot", "region": "korea", "tagline": "Unified mobility OS for autonomous Hyundai/Kia vehicles.", "description": "UMOS is 42dot's unified mobility OS connecting AVs, rider apps, fleet management, and city infrastructure for Hyundai Motor Group.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Hyundai-owned; ships in Hyundai/Kia SDV roadmap", "technicalImpact": 77, "competitors": [ "Waymo Driver", "Mobileye Drive", "Tesla SDV", "VW.OS" ], "features": [ "Autonomous mobility OS", "Fleet management", "SDV platform", "Hyundai/Kia integration" ], "pricing": { "free": "N/A", "pro": "Enterprise only", "enterprise": "Contact sales" }, "link": "https://42dot.ai", "apiLink": null, "aiCoreShare": 78, "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 80 -> 77. Hyundai-backed SDV platform with a real but small Seoul autonomous shuttle pilot; no fleet scale, program economics or external customers disclosed." }, { "id": "42dot-sdv-platform", "name": "42dot SDV Platform", "company": "42dot", "region": "korea", "tagline": "Software-defined vehicle platform for Hyundai Motor Group.", "description": "42dot's SDV platform centralizes OS, middleware, and AI workloads for next-generation Hyundai and Kia software-defined vehicles.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Hyundai Motor Group standard SDV stack", "technicalImpact": 78, "competitors": [ "Tesla SDV", "VW.OS", "Sonatus", "Aptiv SVA" ], "features": [ "SDV middleware", "OTA updates", "Vehicle AI workloads", "OEM integration" ], "pricing": { "free": "N/A", "pro": "OEM only", "enterprise": "Contact sales" }, "link": "https://42dot.ai", "apiLink": null, "aiCoreShare": 70 }, { "id": "a2z-av-stack", "name": "a2z AV Stack", "company": "Autonomous a2z", "region": "korea", "tagline": "Level 4 autonomous-driving stack for shuttles, robotaxis, and AV trucks.", "description": "Autonomous a2z' AV stack covers perception, prediction, planning, and control for level-4 autonomous services across multiple Korean cities.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Korea's most active independent AV-permit holder", "technicalImpact": 75, "competitors": [ "Waymo", "42dot", "Pony.ai", "Mobileye Drive" ], "features": [ "L4 autonomy", "Shuttle / robotaxi / truck", "Korean city deployments", "OEM-agnostic" ], "pricing": { "free": "N/A", "pro": "Pilot agreements", "enterprise": "Contact sales" }, "link": "https://autoa2z.ai", "apiLink": null, "aiCoreShare": 86 }, { "id": "wirobotics-wim", "name": "WIM", "nativeName": "윔", "company": "WIRobotics", "region": "korea", "tagline": "Wearable walking-assist robot generating real-world movement data.", "description": "WIM is WIRobotics' wearable walking-assist robot used for mobility and rehabilitation, collecting the human-movement data that feeds the ALLEX humanoid platform.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Production deployments in Korean rehab and senior-care channels", "technicalImpact": 72, "competitors": [ "Honda Walking Assist", "ReWalk", "Ekso Bionics", "Cyberdyne HAL" ], "features": [ "Wearable exo-actuation", "Gait analysis", "Rehabilitation use", "Movement dataset" ], "pricing": { "free": "N/A", "pro": "Hardware purchase", "enterprise": "Contact sales" }, "link": "https://wirobotics.com", "apiLink": null, "aiCoreShare": 70 }, { "id": "wirobotics-allex", "name": "ALLEX", "company": "WIRobotics", "region": "korea", "tagline": "Korean humanoid robot trained on real-world human movement data.", "description": "ALLEX is WIRobotics' humanoid platform. Selected for the 2026 AWS + NVIDIA Physical AI Fellowship.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "AWS + NVIDIA Physical AI Fellowship 2026", "technicalImpact": 76, "competitors": [ "figure", "1X NEO", "Tesla Optimus", "Unitree H1" ], "features": [ "Humanoid platform", "Human-movement dataset", "Whole-body control", "Manipulation" ], "pricing": { "free": "N/A", "pro": "Pilot program", "enterprise": "Contact sales" }, "link": "https://wirobotics.com", "apiLink": null, "aiCoreShare": 90 }, { "id": "holiday-humanoid", "name": "Holiday Humanoid", "company": "Holiday Robotics", "region": "korea", "tagline": "Early-stage Korean general-purpose humanoid platform.", "description": "Holiday Robotics' humanoid platform is in active development as a Korean entrant in the global general-purpose humanoid race.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Pre-product; backed by Korean deep-tech VCs", "technicalImpact": 62, "competitors": [ "figure", "1X NEO", "WIRobotics ALLEX", "Unitree H1" ], "features": [ "Humanoid platform", "General-purpose tasks", "AI-driven control" ], "pricing": { "free": "N/A", "pro": "N/A", "enterprise": "Contact sales" }, "link": "https://holiday-robotics.com", "apiLink": null, "aiCoreShare": 88 }, { "id": "nearthlab-autonomous-drone-inspection", "name": "Nearthlab Autonomous Drone Inspection", "company": "Nearthlab", "region": "korea", "tagline": "Autonomous drone AI for wind, energy, and infrastructure inspection.", "description": "Nearthlab's autonomous-flight platform combines deep-learning vision with hardware-agnostic drone control for wind turbines, energy assets, and industrial infrastructure.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Named wind-turbine operator customers in the US, Europe, and Asia", "technicalImpact": 77, "competitors": [ "Skydio", "Percepto", "DJI Matrice", "Anduril Ghost" ], "features": [ "Autonomous flight", "Defect detection", "Wind-turbine inspection", "Industrial deployment" ], "pricing": { "free": "N/A", "pro": "Subscription", "enterprise": "Contact sales" }, "link": "https://nearthlab.com", "apiLink": null, "aiCoreShare": 84, "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 80 -> 77. Historical blade-inspection customers are press-documented, but this row discloses no current customers, fleet size or usage and carries no sources." }, { "id": "nearthlab-defense-drone", "name": "Nearthlab Defense Drone", "company": "Nearthlab", "region": "korea", "tagline": "Autonomous defense drone built on Nearthlab's industrial-inspection stack.", "description": "Defense-focused Nearthlab drone for tactical reconnaissance and target identification.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Korean defense-evaluation pilots", "technicalImpact": 73, "competitors": [ "Skydio X10D", "Anduril Ghost", "Shield AI V-BAT", "Parrot Anafi" ], "features": [ "Tactical autonomy", "Vision perception", "Long-endurance flight" ], "pricing": { "free": "N/A", "pro": "Government contract", "enterprise": "Contact sales" }, "link": "https://nearthlab.com", "apiLink": null, "aiCoreShare": 86 }, { "id": "beyond-honeycomb-grill-x", "name": "GRILL X", "company": "Beyond Honeycomb", "region": "korea", "tagline": "AI cooking robot that replicates chef-grade grilling with real-time sensing.", "description": "GRILL X is Beyond Honeycomb's AI cooking robot grilling meats with chef-grade results.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "$7.2M Series A bridge in 2025-07", "technicalImpact": 68, "competitors": [ "Miso Robotics", "Picnic", "Karakuri", "Chef Robotics" ], "features": [ "AI cooking control", "Vision + thermal sensing", "Recipe replication", "Restaurant deployment" ], "pricing": { "free": "N/A", "pro": "Hardware + subscription", "enterprise": "Contact sales" }, "link": "https://www.beyondhoneycomb.com", "apiLink": null, "aiCoreShare": 82 }, { "id": "seoul-robotics-sensr", "name": "SENSR", "company": "Seoul Robotics", "region": "korea", "tagline": "3D perception software that turns infrastructure LiDAR into AI insights.", "description": "SENSR is Seoul Robotics' 3D perception platform on infrastructure-mounted LiDAR for smart-city and AV applications.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Used by BMW Group, smart-city deployments", "technicalImpact": 78, "competitors": [ "Velodyne Vella", "Outsight", "Innoviz", "Aeva" ], "features": [ "3D perception", "LiDAR fusion", "Infrastructure-mounted", "Multi-sensor support" ], "pricing": { "free": "N/A", "pro": "Subscription", "enterprise": "Contact sales" }, "link": "https://www.seoulrobotics.org", "apiLink": null, "aiCoreShare": 86 }, { "id": "seoul-robotics-level-5", "name": "Level 5 Control Tower", "company": "Seoul Robotics", "region": "korea", "tagline": "Infrastructure-based autonomous-vehicle control system.", "description": "Uses infrastructure LiDAR to autonomously move vehicles inside enclosed environments — production deployment at BMW Group.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "BMW Group production deployment for plant logistics", "technicalImpact": 78, "competitors": [ "Waymo Driver", "Mobileye Drive", "Outsight", "Innoviz" ], "features": [ "Infrastructure autonomy", "Plant logistics", "Multi-vehicle coordination", "L4 in restricted areas" ], "pricing": { "free": "N/A", "pro": "Enterprise license", "enterprise": "Contact sales" }, "link": "https://www.seoulrobotics.org", "apiLink": null, "aiCoreShare": 84, "scoringRationale": "2026-07-29 Korean top-tier stress test: reduced 80 -> 78. The BMW plant deployment of LV5 CTRL TWR is press-documented — a named global OEM production use that clears the unnamed tier — but the row cites no sources and discloses no scale. Needs source URLs to hold 78." }, { "id": "robotis-dynamixel", "name": "Dynamixel", "company": "ROBOTIS", "region": "korea", "tagline": "Programmable smart actuators used in robotics research worldwide.", "description": "Networked, programmable servomotors used by global robotics labs and humanoid developers.", "users": "Global research labs", "userCount": 0, "adoptionSignal": "Standard servo in academic robotics", "technicalImpact": 76, "competitors": [ "Maxon Motor", "Harmonic Drive", "Nabtesco", "Bota Systems" ], "features": [ "Smart actuators", "Network bus", "Position/velocity/torque control", "Research standard" ], "pricing": { "free": "N/A", "pro": "Per unit", "enterprise": "Contact sales" }, "link": "https://www.robotis.us/", "apiLink": null, "aiCoreShare": 60 }, { "id": "robotis-turtlebot", "name": "TurtleBot", "company": "ROBOTIS", "region": "korea", "tagline": "Open-source ROS reference mobile robot.", "description": "Open-source ROS mobile-robot reference platform produced by Robotis with Open Robotics.", "users": "Global research and education customers", "userCount": 0, "adoptionSignal": "Reference platform for ROS / Open Robotics", "technicalImpact": 72, "competitors": [ "Clearpath", "iRobot Create", "Boston Dynamics Spot", "Unitree Go2" ], "features": [ "ROS-native", "Mobile robotics", "Educational use", "Customizable" ], "pricing": { "free": "Open source", "pro": "Per unit", "enterprise": "Contact sales" }, "link": "https://www.robotis.us/", "apiLink": null, "aiCoreShare": 64 }, { "id": "robotis-gaemi", "name": "GAEMI", "company": "ROBOTIS", "region": "korea", "tagline": "Autonomous indoor delivery robot.", "description": "Robotis' autonomous indoor delivery robot for hospitality, hospital, and office environments.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Korean hospital and hotel pilots", "technicalImpact": 65, "competitors": [ "Bear Robotics", "Pudu Robotics", "Starship", "Nuro" ], "features": [ "Indoor autonomy", "SLAM navigation", "Multi-floor capable", "Delivery payload" ], "pricing": { "free": "N/A", "pro": "Hardware + subscription", "enterprise": "Contact sales" }, "link": "https://www.robotis.us/", "apiLink": null, "aiCoreShare": 70 }, { "id": "mobilint-aries", "name": "ARIES", "company": "Mobilint", "region": "korea", "tagline": "Edge-AI NPU for industrial vision and robotics.", "description": "ARIES is Mobilint's high-performance edge NPU for industrial vision, robotics, smart infrastructure, and embedded AI workloads.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Korean industrial-vision and AI-camera partners", "technicalImpact": 70, "competitors": [ "Hailo-8", "NVIDIA Jetson", "Ambarella", "Untether AI" ], "features": [ "Edge AI NPU", "Industrial vision", "Robotics", "Low-power inference" ], "pricing": { "free": "N/A", "pro": "Per chip", "enterprise": "Contact sales" }, "link": "https://www.mobilint.com", "apiLink": null, "aiCoreShare": 86 }, { "id": "mobilint-mlx-a1", "name": "MLX-A1", "company": "Mobilint", "region": "korea", "tagline": "Industrial AI PC with 80 TOPS NPU.", "description": "Mobilint's industrial AI PC packaging an 80 TOPS NPU for vision, voice, and on-device LLM tasks.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Industrial deployments", "technicalImpact": 68, "competitors": [ "NVIDIA Jetson AGX", "Lenovo Edge", "Dell Edge Gateway", "Hailo Edge" ], "features": [ "AI PC", "80 TOPS NPU", "Edge deployment", "Vision + voice + LLM" ], "pricing": { "free": "N/A", "pro": "Per unit", "enterprise": "Contact sales" }, "link": "https://www.mobilint.com", "apiLink": null, "aiCoreShare": 80 }, { "id": "bos-eagle-n", "name": "Eagle-N", "company": "Bos Semiconductors", "region": "korea", "tagline": "Automotive AI accelerator for ADAS and cockpit AI.", "description": "Bos Semiconductors' automotive AI accelerator for ADAS, cockpit AI, and privacy-first on-device processing inside vehicles.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "OEM evaluation engagements", "technicalImpact": 68, "competitors": [ "Mobileye EyeQ", "Ambarella CV", "NVIDIA Drive", "Qualcomm Ride" ], "features": [ "Automotive AI", "ADAS perception", "Cockpit AI", "On-device privacy" ], "pricing": { "free": "N/A", "pro": "Per chip", "enterprise": "Contact sales" }, "link": "https://www.bos-semi.com", "apiLink": null, "aiCoreShare": 88 }, { "id": "deep-fusion-rapa", "name": "RAPA 4D Imaging Radar", "company": "Deep Fusion AI", "region": "korea", "tagline": "4D imaging radar with LiDAR-equivalent perception.", "description": "Deep Fusion AI's 4D imaging-radar product delivering LiDAR-equivalent perception at lower cost. CES 2026 Best of Innovation.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "CES 2026 Best of Innovation", "technicalImpact": 70, "competitors": [ "Arbe Robotics", "Continental ARS", "Aeva Aeries", "Mobileye Imaging Radar" ], "features": [ "4D imaging radar", "Deep-learning post-processing", "ADAS perception", "Lower cost than LiDAR" ], "pricing": { "free": "N/A", "pro": "OEM only", "enterprise": "Contact sales" }, "link": "https://www.deep-fusion.com", "apiLink": null, "aiCoreShare": 82 }, { "id": "stradvision-svnet", "name": "SVNet", "company": "StradVision", "region": "korea", "tagline": "Camera-perception software for automotive ADAS production programs", "description": "SVNet is STRADVISION's camera-perception software for automotive ADAS, detecting vehicles, pedestrians and lanes on in-vehicle hardware. The company reported more than 5M cumulative vehicle deployments in April 2026. This is production use in a specialized automotive market, not a general-purpose assistant or 5M active software customers.", "users": "5M+ cumulative vehicle deployments (company-reported, Apr 2026)", "userCount": 5000000, "adoptionSignal": "IPO filing discloses 5M+ vehicles, 13 OEM/Tier-1 programs, 50+ models; strategic investors Aptiv, Hyundai Motor, LG Electronics", "technicalImpact": 84, "aiCoreShare": 92, "competitors": [ "Mobileye EyeQ", "Qualcomm Arriver", "NVIDIA DRIVE", "Phantom AI", "a2z-av-stack" ], "features": [ "Camera perception for FCW, AEB, LDW, LKA, ACC, HDA", "L0-L2+ tiers including 8MP long-range and AR variants", "Embedded on 18-30+ automotive SoC platforms", "SurroundVision parking and low-speed perception" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://stradvision.com/dh/product_front", "apiLink": null, "sourceUrls": [ "https://stradvision.com/dh/product_front", "https://www.prnewswire.com/news-releases/stradvision-files-for-kosdaq-ipo-targets-offering-of-7-million-shares-302738810.html", "https://www.ajupress.com/view/20260630095270188" ], "scoringRationale": "Reviewed 2026-09-29: retains Impact 84 for disclosed multi-OEM production deployment, not IPO status or the importance of autonomous driving. Its buyer market is specialized; installed vehicles are not active users or current market share." }, { "id": "neuromeka-indy", "name": "Indy", "company": "Neuromeka", "region": "korea", "tagline": "Korean collaborative robots with AI vision", "description": "Indy is Korea's first domestically designed collaborative-robot line, using sensorless collision detection and impedance-control direct teaching, programmed from an Android tablet pendant with IndyDCP/gRPC APIs and a Python SDK. Pro variants integrate IndyEye deep-learning vision for welding and picking. Kyochon F&B runs ~40 frying robots across ~30 branches with an LA rollout underway, and Neuromeka claims Korea's #1 2024 cobot revenue; FY2024 revenue was KRW 25.3B (+84% YoY).", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Kyochon F&B ~40 robots across ~30 branches with LA expansion; NSF certification for US foodservice; company-claimed #1 2024 Korean cobot revenue; FY2024 revenue KRW 25.3B", "technicalImpact": 76, "aiCoreShare": 82, "competitors": [ "Universal Robots", "doosan-robot-lab-ai", "Rainbow Robotics RB Series", "Techman Robot" ], "features": [ "Sensorless collision detection for worker safety", "Impedance-control direct teaching", "IndyEye deep-learning vision integration", "IndyDCP/gRPC APIs and Python SDK" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://en.neuromeka.com/indy", "apiLink": "https://docs.neuromeka.com", "sourceUrls": [ "https://en.neuromeka.com/indy", "https://www.theworldfolio.com/news/beyond-robotics-how-neuromeka-is-scaling-physical-ai-across-industries/5415/", "https://www.globenewswire.com/news-release/2023/08/29/2733849/0/en/Neuromeka-Attains-NSF-Certification-and-Forges-Strategic-Partnerships-for-U-S-Market-Entry-with-Robotic-Frying-Solution.html" ], "scoringRationale": "76 reflects a listed domestic cobot leader with named foodservice deployment, US certification, and fast revenue growth; held below 80 because fleet counts are press-reported and global share remains small.", "nativeName": "인디" }, { "id": "openedges-enlight-npu", "name": "ENLIGHT NPU IP", "company": "OPENEDGES Technology", "region": "korea", "tagline": "Licensable Korean NPU IP for edge-AI chips", "description": "ENLIGHT is licensable neural-processing-unit IP for chip designers: ENLIGHT Classic for 4/8-bit mixed-precision edge inference and ENLIGHT Pro for transformer/LLM workloads at 4096 MACs/cycle, scaling from 8 to hundreds of TOPS. It ships as RTL with a compiler toolchain (ONNX, TensorFlow Lite, Darknet) and is licensed alongside OPENEDGES' ORBIT memory-subsystem IP. Named licensees include Telechips (automotive SoCs) and Eyenix (AI surveillance chipsets); the company states 40+ customers and 80+ IP integrations across its lines.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Named licensees Telechips and Eyenix; company-stated 40+ customers / 80+ IP integrations across IP lines; Korea's only listed semiconductor-IP vendor", "technicalImpact": 76, "aiCoreShare": 88, "competitors": [ "Arm Ethos", "Ceva NeuPro", "Synopsys ARC NPX", "Cadence Neo NPU" ], "features": [ "Mixed-precision edge-inference NPU IP (Classic)", "Transformer/LLM-capable ENLIGHT Pro to hundreds of TOPS", "DRAM-traffic-optimized layer partitioning", "ONNX/TFLite/Darknet compiler toolchain" ], "pricing": { "free": "N/A", "pro": "IP license", "enterprise": "Contact sales" }, "link": "https://www.openedges.com/npu", "apiLink": null, "sourceUrls": [ "https://www.openedges.com/npu", "https://www.design-reuse.com/news/14604-openedges-and-telechips-collaborates-for-automotive-applications/", "https://www.design-reuse.com/news/9014-openedges-ai-accelerator-npu-memory-subsystem-ip-licensed-for-eyenix-ai-powered-surveillance-camera-chipset-/" ], "scoringRationale": "76 reflects Korea's only listed chip-IP vendor with named NPU licensees shipping silicon; held below 80 because royalty volumes are undisclosed and the IP competes with far larger Arm/Synopsys/Cadence portfolios." }, { "id": "clobot-croms", "name": "CROMS", "company": "Clobot", "region": "korea", "tagline": "Heterogeneous robot fleet-control platform", "description": "CROMS lets one operator control robots from different manufacturers and movement types in a single cloud system, providing collision-free multi-robot path optimization, real-time monitoring, task scheduling, and automated reports, with robots connected through Clobot's CRCS command spec and ROS1/ROS2/non-ROS adapters. KED Global reports 130+ company-level clients including Hyundai Motor and Boston Dynamics. CROMS-specific license counts are not split out.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "130+ company-level clients incl. Hyundai Motor and Boston Dynamics (KED Global); KOSDAQ listing Oct 2024", "technicalImpact": 76, "aiCoreShare": 88, "competitors": [ "InOrbit", "Formant", "Rapyuta Robotics", "Meili FMS" ], "features": [ "Multi-vendor heterogeneous fleet control", "Collision-free multi-robot path optimization", "Real-time monitoring and task scheduling", "CRCS interface for ROS1/ROS2/non-ROS robots" ], "pricing": { "free": "Free tier", "pro": "Subscription", "enterprise": "Contact sales" }, "link": "https://www.clobot.co.kr/croms", "apiLink": null, "sourceUrls": [ "https://www.clobot.co.kr/croms", "https://www.kedglobal.com/robotics/newsView/ked202408270008" ], "scoringRationale": "76 reflects Korea's leading vendor-neutral robot orchestration layer with 130+ reported clients and a public listing; held below 80 because platform-specific license counts and recurring revenue are not disclosed.", "nativeName": "크롬스" }, { "id": "clobot-chameleon", "name": "CHAMELEON", "company": "Clobot", "region": "korea", "tagline": "Hardware-agnostic robot navigation stack", "description": "CHAMELEON adds mapping, localization, and autonomous driving intelligence to third-party robot platforms — wheeled or quadruped — using 3D LiDAR SLAM with multi-sensor fusion, supporting multiple heterogeneous robots on one unified map and integrating with CROMS for fleet operation. Deployments include Boston Dynamics SPOT-based patrol robots. Standalone adoption counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Documented deployments incl. Boston Dynamics SPOT-based patrol projects; sold alongside CROMS", "technicalImpact": 70, "aiCoreShare": 88, "competitors": [ "SLAMcore", "Nav2 (open source)", "seoul-robotics-sensr", "Kudan SLAM" ], "features": [ "3D LiDAR SLAM precision mapping", "Multi-sensor fusion localization (LiDAR, camera, IMU)", "Wheeled and quadruped platform support", "Predictive collision avoidance" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://www.clobot.co.kr/Chameleon", "apiLink": null, "sourceUrls": [ "https://www.clobot.co.kr/Chameleon", "https://www.kedglobal.com/robotics/newsView/ked202408270008" ], "scoringRationale": "70 reflects a credible navigation layer that extends Clobot's orchestration into third-party hardware; kept below CROMS because standalone adoption is undisclosed.", "nativeName": "카멜레온" }, { "id": "uvify-ifo", "name": "IFO", "company": "UVify", "region": "korea", "tagline": "Swarm light-show drone flown in fleets of thousands", "description": "IFO is UVify's flagship swarm drone for choreographed light shows: a 990g airframe with up to 25 minutes of flight, RTK-GNSS hovering at ±0.1m accuracy, quad-constellation navigation, and a 24-bit WRGB LED array visible beyond 7km, operated in fleets up to thousands of units. The product page reports 40K+ IFO-family drones deployed across 25+ countries, with most revenue generated overseas. UVify raised a record KRW 60B round in early 2026 and took Naver strategic investment for defense expansion.", "users": "40K+ drones deployed (company-stated)", "userCount": 40000, "adoptionSignal": "Product-page deployment counters: 40K+ drones across 25+ countries; largest-ever Korean drone investment round (KRW 60B, Feb 2026); Naver defense-AX investment (Jun 2026)", "technicalImpact": 78, "aiCoreShare": 82, "competitors": [ "HighGreat", "Damoda", "Verge Aero", "Pablo Air" ], "features": [ "Swarm choreography with thousands of units", "RTK-GNSS ±0.1m hovering accuracy", "24-bit WRGB LED visible beyond 7km", "25-minute flight, sealed maintenance-free motors" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://www.uvify.com/product/ifo", "apiLink": null, "sourceUrls": [ "https://www.uvify.com/product/ifo", "https://en.sedaily.com/finance/2026/03/04/drone-startup-uvify-raises-42m-in-fresh-funding", "https://en.sedaily.com/technology/2026/06/01/naver-bets-on-defense-ax-with-drone-startup-uvify-investment" ], "scoringRationale": "78 reflects quantified global deployment (40K+ units, 25+ countries) of exported Korean autonomous hardware with record capital backing; held below 80 because figures are company-stated and defense products are not yet publicly cataloged." }, { "id": "hyperaccel-bertha-500", "name": "Bertha 500", "company": "HyperAccel", "region": "korea", "tagline": "4nm LLM-inference accelerator on LPU architecture", "description": "Bertha 500 is HyperAccel's 4nm datacenter accelerator built on the Latency Processing Unit (LPU) architecture, with dataflow matched to autoregressive token generation: 768 TFLOPS FP8, 128-256GB LPDDR5X at 546GB/s, 256MB SRAM, PCIe Gen5, built-in networking for rack-scale inference, and a 250W TDP, running PyTorch/ONNX/vLLM through the HyperDex stack. First silicon came back from Samsung Electronics' 4nm foundry service with mass production in preparation. No commercial cloud or OEM deployment has been confirmed; prospective customers remain unannounced.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Pre-mass-production; first 4nm silicon fabricated; Upstage sovereign-AI consortium inference partner; no confirmed named cloud deployment yet", "technicalImpact": 76, "aiCoreShare": 90, "competitors": [ "groq-lpu", "atom-max", "furiosa-rngd", "SambaNova" ], "features": [ "LPU dataflow architecture with ~90% utilization across batch sizes", "4nm, 768 TFLOPS FP8, 128-256GB LPDDR5X", "Built-in networking for rack-scale inference", "HyperDex software: PyTorch, ONNX, vLLM" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://hyperaccel.ai/ha_product/bertha-500/", "apiLink": null, "sourceUrls": [ "https://hyperaccel.ai/ha_product/bertha-500/", "https://www.thelec.net/news/articleView.html?idxno=5787", "https://hyperaccel.ai/hyperaccel-raises-55-billion-krw-in-series-a-funding-for-llm-focused-ai-semiconductor-development/" ], "scoringRationale": "76 sits under the AI-hardware evidence cap: differentiated inference-first silicon with strong funding and a sovereign-AI consortium role, but userCount 0, no named cloud/OEM deployment, and pre-mass-production status." }, { "id": "morai-sim-drive", "name": "MORAI SIM Drive", "company": "MORAI", "region": "korea", "tagline": "Digital-twin simulation for autonomous driving", "description": "MORAI SIM Drive is a high-fidelity digital-twin platform for developing and validating autonomous-driving systems: HD-map environments with vehicle dynamics and camera/LiDAR/GPS/IMU sensor models, scenario generation from real road data, ISO 26262 ASIL-D certification, vehicle-in-the-loop testing, and cloud runs of thousands of parallel simulations with KITTI/Cityscapes dataset export. MORAI reported 100+ corporate clients including Hyundai Mobis and Hyundai AutoEver at its 2022 Series B; current counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "100+ corporate clients incl. Hyundai Mobis, Hyundai AutoEver (2022 Series B disclosure); NVIDIA/Ansys/dSPACE partnerships", "technicalImpact": 74, "aiCoreShare": 88, "competitors": [ "Applied Intuition", "dSPACE AURELION", "IPG CarMaker", "Foretellix" ], "features": [ "HD-map digital twins with full sensor models", "Scenario generation from real road data", "ISO 26262 ASIL-D certification, vehicle-in-the-loop", "Cloud simulation with thousands of parallel runs" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://www.morai.ai/drive", "apiLink": null, "sourceUrls": [ "https://www.morai.ai/drive", "https://techcrunch.com/2022/02/07/automotive-simulation-platform-morai-wants-to-expand-its-international-footprint-with-20-8m-series-b/" ], "scoringRationale": "74 reflects Korea's leading AV simulation vendor with 100+ disclosed clients and safety certification; held below 80 because the client figure dates to 2022 and global leaders (Applied Intuition) operate at far larger scale." }, { "id": "rideflux-driver", "name": "RideFlux Driver", "company": "RideFlux", "region": "korea", "tagline": "L4 autonomous-driving software in paid commercial service", "description": "RideFlux Driver is a full-stack autonomous-driving system sold in an L4 variant (camera+radar+LiDAR driverless operation with remote monitoring) and an L2+ camera+radar variant, with one hardware-agnostic stack spanning sedans to 25-ton trucks plus in-house HD mapping and remote operations. It received Korea's first paid autonomous freight permit (MOLIT, April 2026) and launched paid autonomous trucking with Hanjin on a 116km Gunsan-Jeonju-Daejeon route on July 1, 2026, alongside Jeju robotaxi/shuttle services with Kakao Mobility. Fleet size remains small while the company prepares an H2 2026 KOSDAQ listing.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Korea's first paid autonomous freight permit (Apr 2026); Hanjin paid trucking service launched Jul 1, 2026; Kakao Mobility NEMO Ride robotaxi on Jeju; temporary L4 permit Jun 2024", "technicalImpact": 76, "aiCoreShare": 92, "competitors": [ "a2z-av-stack", "42dot-umos", "Kodiak Robotics", "Mars Auto" ], "features": [ "L4 driverless operation with 360-degree multi-sensor perception", "L2+ camera+radar consumer variant", "Single stack from sedans to 25-ton trucks", "In-house HD mapping and remote operations" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://www.rideflux.com/solution", "apiLink": null, "sourceUrls": [ "https://www.rideflux.com/solution", "https://en.sedaily.com/technology/2026/07/01/rideflux-hanjin-launch-koreas-first-paid-autonomous-truck", "https://en.sedaily.com/finance/2026/04/16/rideflux-becomes-first-in-korea-to-receive-permit-for-paid" ], "scoringRationale": "76 reflects nation-first regulatory milestones and revenue-generating driverless services with named partners (Hanjin, Kakao Mobility); held below 80 because the deployed fleet is small and route economics are unproven." }, { "id": "config-loop", "name": "Loop", "company": "Config Intelligence", "region": "korea", "tagline": "Human-demonstration-to-robot-policy platform", "description": "Loop converts real-world human and teleoperation demonstrations into deployable robot policies across six stages — Collect, Curate, Convert, Train, Deploy, Evaluate — powered by GPU-based conversion and policy engines, supporting arms like Franka Research 3 and Rainbow Robotics RB-Y1 with sub-50ms deployed-policy latency. Config runs ~300 staff producing data in Seoul and Hanoi and has accumulated 100K+ hours of human motion data; TechCrunch reports unnamed large-manufacturer, integrator, agriculture, and defense customers. Customer names and revenue are not disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "100K+ hours of proprietary human motion data; unnamed manufacturer/integrator customers (TechCrunch); $27M seed at $200M+ valuation backed by Samsung/Hyundai/LG/SK venture arms", "technicalImpact": 76, "aiCoreShare": 90, "competitors": [ "Scale AI", "Physical Intelligence", "NVIDIA Isaac GR00T", "AgiBot World" ], "features": [ "Multi-camera demonstration capture with AI-assisted curation", "Human-to-robot trajectory conversion engine", "Fine-tuning of CFG-1 or open robot foundation models", "Sub-50ms on-robot policy deployment with evaluation loop" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://config.inc/products/loop", "apiLink": null, "sourceUrls": [ "https://config.inc/products/loop", "https://techcrunch.com/2026/05/11/koreas-biggest-manufacturers-back-config-the-tsmc-of-robot-data/" ], "scoringRationale": "76 reflects a differentiated robot-data platform with a proprietary 100K-hour corpus, four-chaebol backing, and reported paying customers; held below 80 because customers are unnamed and the company is barely a year old." }, { "id": "config-cfg-1", "name": "CFG-1", "company": "Config Intelligence", "region": "korea", "tagline": "Bimanual robot manipulation foundation model", "description": "CFG-1 is Config's robot foundation model for general-purpose bimanual manipulation, trained on the company's 100K+ hours of proprietary human motion data and offered as the default trainable base model inside the Loop platform, where customers fine-tune it into task-specific policies deployable on supported arms. No standalone adoption figures are disclosed, and the robot-foundation-model category remains pre-standardization.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Live model page; trained on 100K+ hours of proprietary motion data; no standalone adoption disclosed", "technicalImpact": 72, "aiCoreShare": 95, "competitors": [ "Physical Intelligence pi-0", "NVIDIA Isaac GR00T", "rldx-1", "Figure Helix" ], "features": [ "Bimanual manipulation foundation model", "Trained on 100K+ hours of proprietary human motion data", "Fine-tunable via the Loop platform", "Deployable on supported robot arms" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://config.inc/model", "apiLink": null, "sourceUrls": [ "https://config.inc/model", "https://techcrunch.com/2026/05/11/koreas-biggest-manufacturers-back-config-the-tsmc-of-robot-data/" ], "scoringRationale": "72 reflects a proprietary-data-advantaged robot foundation model in a strategically hot category; kept below Loop because model-specific adoption, benchmarks, and deployments are not yet public." }, { "id": "bear-servi-plus", "name": "Servi Plus", "company": "Bear Robotics", "region": "korea", "tagline": "Autonomous food-service robot at chain scale", "description": "Servi Plus is Bear Robotics' flagship food-service robot for restaurants, hotels, and senior living: an 88lb/16-entree payload across four trays, autonomous navigation with 360-degree obstacle detection and real-time rerouting, active suspension for spill-free delivery, and multi-robot orchestration. The fleet — several thousand robots at chains including Denny's and Chili's across the US, Korea, and Japan — now anchors LG Electronics' commercial robot business (LG holds 51%), with the company site reporting ~107M cumulative deliveries as of Q1 2026. Per-model fleet counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "~107M cumulative deliveries reported on first-party site (Q1 2026); several thousand robots incl. Denny's/Chili's (TechCrunch); US/Korea/Japan markets (LG newsroom)", "technicalImpact": 78, "aiCoreShare": 82, "competitors": [ "Keenon Robotics", "Pudu Robotics", "Richtech Robotics", "OrionStar" ], "features": [ "88lb / 16-entree payload with modular trays", "Autonomous navigation with 360-degree obstacle detection", "Active suspension for spill-free delivery", "Multi-robot orchestration with LED branding display" ], "pricing": { "free": "N/A", "pro": "Per deployment", "enterprise": "Contact sales" }, "link": "https://www.bearrobotics.ai/servi-plus", "apiLink": null, "sourceUrls": [ "https://www.bearrobotics.ai/servi-plus", "https://www.lg.com/global/newsroom/news/corporate/lg-acquires-majority-stake-in-bear-robotics-to-bolster-robotics-capabilities/", "https://techcrunch.com/2024/03/12/bear-robotics-a-robot-waiter-startup-just-picked-up-60m-from-lg/" ], "scoringRationale": "78 reflects chain-scale named deployments, first-party cumulative delivery metrics, and LG's majority acquisition as commercial validation; held below 80 because current fleet size and unit economics are not disclosed.", "nativeName": "서비 플러스" }, { "id": "mech-mind-vision-suite", "name": "Mech-Vision & Mech-Viz", "company": "Mech-Mind Robotics", "region": "china", "tagline": "3D machine vision and robot-planning suite for industrial automation", "description": "Mech-Vision & Mech-Viz combine 3D machine vision, deep-learning recognition, robot communication, and motion planning for bin picking, machine tending, depalletizing, assembly, and inspection. The suite is sold with Mech-Eye cameras and other standardized components rather than as an internal system.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2018", "technicalImpact": 82, "competitors": [ "cognex-visionpro-deep-learning", "keyence-vision-system", "siemens-industrial-copilot", "Intrinsic Flowstate" ], "features": [ "No-code 3D machine-vision workflows", "2D and 3D deep-learning recognition", "Robot path planning and communication", "More than 1,000 supported robot models", "Industrial bin picking, tending and inspection" ], "pricing": { "enterprise": "Contact sales" }, "link": "https://www.mech-mind.com/products/mech-vision-machine-vision-software.html", "apiLink": "https://downloads.mech-mind.com/?tab=tab-suite", "aiCoreShare": 86, "sourceUrls": [ "https://www.mech-mind.com/products/mech-vision-machine-vision-software.html", "https://downloads.mech-mind.com/?tab=tab-suite", "https://www.hkex.com.hk/eng/market/sec_tradinfo/tradarng/tradarng_news/currentmonth/e9615a_260828.htm" ], "adoptionSignal": "Mech-Mind's IPO materials describe commercial deployment across automotive, new energy, consumer electronics, logistics and general manufacturing; the company sells in nearly 50 countries, but suite-specific installations are not disclosed.", "scoringRationale": "Impact 82 reflects a commercially shipped physical-AI component suite with multi-industry and international reach. It remains below the 86 tier because customer and installation counts are reported at company or portfolio level rather than for this suite alone.", "versions": [ { "name": "Mech-Vision & Mech-Viz 2.2.1", "released": "2026-07" }, { "name": "Mech-Vision & Mech-Viz 2.1.2", "released": "2025-05" } ], "versionsUpdated": "2026-09-02" }, { "id": "seer-robot-brain", "name": "SEER Robot Brain", "company": "SEER Robotics", "region": "china", "tagline": "Robot controllers and fleet software for industrial mobile robots", "description": "SEER Robot Brain is the controller-and-software platform behind the company's industrial mobile robots. It combines SRC controller hardware, navigation and motion control, the RDS robot-management system, and the M4 logistics platform for factories and warehouses; SEER also sells forklifts, lifting robots, bin-handling robots, and delivery robots built on the stack.", "users": "2,500+ company customers (portfolio-level)", "userCount": 0, "releaseDate": "2015", "technicalImpact": 84, "competitors": [ "mech-mind-vision-suite", "Geek+", "Hai Robotics", "Quicktron" ], "features": [ "SRC industrial robot controllers", "RDS multi-robot management", "M4 smart-logistics orchestration", "Navigation, motion control and safety", "Support for more than 1,000 robot models" ], "pricing": { "enterprise": "Contact sales" }, "link": "https://cn-cdn.seer-group.com/en/robot-build", "apiLink": null, "aiCoreShare": 82, "sourceUrls": [ "https://cn-cdn.seer-group.com/en/robot-build", "https://cn-cdn.seer-group.com/en/media/333", "https://www1.hkexnews.hk/search/titlesearch.xhtml?category=0&lang=EN&market=SEHK&stockId=1000307107" ], "adoptionSignal": "SEER reports 2,500+ company customers worldwide, support for more than 1,000 robot models, and 24.8% global unit share in intelligent robot controllers in 2025; these are company- and portfolio-level figures rather than active users of one software surface.", "scoringRationale": "Impact 84 reflects a quantified global customer base, broad robot compatibility, and commercially deployed controller and fleet-management infrastructure. userCount remains zero because the disclosed 2,500+ figure counts company customers across SEER's portfolio, not end users of this row.", "nativeName": "仙工机器人大脑" } ]