Korea AI Next Waves · July 2026
What Comes After
All This AI?
The AI industry moves in waves: chips, then models, then agents. Each wave crowns different winners, and a country's position is set less by ambition than by structure. This is the timeline view — which wave is cresting, which is next, and where Korea actually stands in each.
Memory was Korea's way into the first wave. Robots are its way into the last one. The brain, in between, belongs to San Francisco — and increasingly to open weights from China.
Wave 1 · 2016 —
Compute & Memory
The physical substrate. Every model trains and serves on accelerators fed by high-bandwidth memory, and supply of both became the binding constraint of the entire industry. Value pooled at the bottleneck: NVIDIA, TSMC, and the HBM duopoly.
Who leads
NVIDIA (~$4.7T), TSMC — with Korean memory inside nearly every accelerator shipped.
Korea's position · Won — but it's the helper layer
Korea's strongest wave. Samsung and SK Hynix control most of global HBM supply; Rebellions and FuriosaAI are the sovereign inference-chip bet, both now pre-IPO. But memory is the helper, not the brain — the value captured here is real yet structurally capped below the model layer's.
Wave 2 · 2020 —
Frontier Models
The brain. Training runs crossed $10B/year and the frontier consolidated to a handful of US labs — Anthropic ($965B post-Series H, S-1 filed) and OpenAI ($852B) above all. China answered by commoditizing: DeepSeek and Qwen made frontier-adjacent intelligence free and open-weight, betting that if the brain becomes cheap, distribution wins.
Who leads
US: Anthropic, OpenAI, Google DeepMind. China: DeepSeek, Qwen, Zhipu (~$102B listed), Moonshot ($20B).
Korea's position · Missed — hedged, not contending
No Korean frontier lab, and the capital math says there won't be one. HyperCLOVA X, EXAONE, Solar, and A.X K1 are sovereign hedges — strategically rational for Korean-language enterprise and data sovereignty, but not global contenders. The honest read: this wave was missed, and chasing it now buys depreciation, not position.
Wave 3 · 2024 — now
Agentic Labor
Models stop being tools and become workers. Software stops being something humans click through and becomes something agents operate — the agent layer eats SaaS, and the economic unit shifts from seats to outcomes. Coding went first because output is verifiable: Claude Code, Codex, and a $29.3B Cursor with a pending ~$60B acquisition are the proof. Finance ops, customer service, legal, and sales are next.
Who leads
US: Claude Code, Codex, Cursor. China: agent apps riding open-weight models at consumer scale.
Korea's position · Contesting the edges
Present, but without a global winner: Liner chases Perplexity from a distance, Allganize builds enterprise agents, Delight.ai ships CX agents, Wrtn owns consumer distribution at home. No Korean Cursor exists — and since developers adopt the best tool regardless of origin, the opening is agent-adjacent: Korean-language enterprise workflows, integrations, and vertical agents rather than horizontal platforms.
Wave 4 · 2026 →
AI for Science
The wave that changes GDP instead of redistributing SaaS revenue. Frontier-lab economics only resolve if models eventually compress R&D itself — drug discovery, materials, mathematics. Early signals are already priced: OpenEvidence at $12B for medical evidence alone, research-grade model products, and AI-discovered candidates entering trials.
Who leads
US: OpenEvidence, frontier-lab research models. Global pharma AI consolidating fast.
Korea's position · Seeds, subscale capital
Seeds exist — Standigm, Galux, Deargen, HITS in drug discovery, and Oncocross became Korea's first AI-drug KOSDAQ listing — but funding is an order of magnitude below US peers. The domain-blind read: Korean bio-AI has scientific credibility and subscale capital. If wave 4 breaks big, Korea participates through partnerships, not platforms.
Wave 5 · 2027 →
Embodiment
The hardware-agnostic brain meets the physical world. Not chips — robots: the moment general models become general controllers, the constraint shifts from intelligence to actuators, batteries, sensors, and the ability to manufacture millions of physical units reliably. That is a manufacturing economy's game.
Who leads
US: Figure, Physical Intelligence ($5.6B), Tesla Optimus. China: Unitree (STAR IPO approved), UBTech (listed). The brain will be American or Chinese — the body is up for grabs.
Korea's position · The structural opening
Korea's real opening, and it is structural: the world's highest robot density per worker, Hyundai with Boston Dynamics, Rainbow Robotics, Doosan, Bear Robotics, Neuromeka, WIRobotics, Holiday Robotics — plus the battery, sensor, and actuator supply chain the whole wave depends on. Korea missed the model wave; it does not have to miss the embodiment wave. Memory was the way into wave 1. Robots are the way into wave 5.