The Osaka AI Factory
Nvidia and the state-backed Japanese AI developer Noetra have signed a binding partnership to construct a national AI factory in Osaka, with 27,500 next-generation Rubin GPUs inside it. Operations are scheduled to begin in 2028. The deal closed on July 16, 2026 at the end of CEO Jensen Huang's two-day Tokyo visit.
The site is pitched as the world's first national AI factory dedicated to physical AI. That positioning — physical rather than conversational — is deliberate. The compute base is meant to train models that drive machines, robots, and industrial systems, not to host chatbots. Nvidia is selling the silicon, the system architecture, and the integration stack into a captive domestic market. That is the play: lock the supply, control the deployment surface.
The Osaka location places the deal inside Japan's manufacturing base. The consortium that will draw on the factory reflects that mix. The deal extends Huang's two-day tour, which also produced chip-material supplier agreements, a robotics coalition, and deepened ties with Toyota, per TechCrunch. The trip's structure suggests a sequenced strategy: secure the compute base, then the application layer.
The Funding Stack
Public money carries the load. The project sits inside METI's FRONTia initiative, backed by a five-year government subsidy of ¥1 trillion, roughly $6.2 billion, per BigGo Finance. Private capital is layered on top: 44 companies are committing alongside the state, including SoftBank, Sony, Honda, NEC, and Toyota.
The funding structure separates compute from consumption. The state pays for the silicon, the integration, and the training runs; the 44-firm consortium functions as the customer base, the validation ground, and the data source. Output is a Japanese-language multimodal foundation model, intended for domestic deployment rather than export. That is a different shape from consumer AI lab funding, where hyperscalers underwrite the compute themselves.
Stated use cases are labor shortages and manufacturing competitiveness. That phrasing is industrial-policy language, not consumer AI. The model's release surface, the licensing terms, and whether the trained weights remain inside Japan are not addressed in the announcements covered by BigGo Finance and Nikkei Asia.
Sovereign AI, Physical AI
The framing Huang used in Tokyo was "Sovereign AI." Data and models stay inside national jurisdiction; the underlying silicon does not have to. That is the conceptual scaffolding for the Osaka deal. The hardware is American, the data and weights are intended to be Japanese, and the access rules are written in Tokyo.
Physical AI is the second pillar. The factory is pitched as a training ground for systems that act on the physical world — factory robots, automation stacks, embodied agents. The inference is that Nvidia is selling into the manufacturing floor by selling into the state. Sovereign-AI compute is the lead product; the 27,500 Rubin GPUs are the moat; the Japanese model is the consumable that the consortium internalizes.
The Osaka deal slots into a pattern of sovereign-AI partnerships Nvidia has struck in Europe and the Middle East, according to Nikkei Asia. The differentiator is the industrial customer base — 44 firms, including automakers and electronics groups, that the state has organized into a single demand signal. The open question is whether the consortium produces models useful enough to justify the ¥1 trillion before the subsidy window closes.
Sources
- techcrunch.com — Jensen Huang Japan Visit: Noetra Sovereign AI
- biggo.com — Jensen Huang Ignites 'Japan AI' in Tokyo: National AI Factory Set for 2028 Launch… (2026-07-17)
- nikkei.com — Nvidia cashes in on sovereign AI, setting sights on Japan ... (2026-07-16)
- co.in — After Jensen Huang's Tokyo Visit, Japan's Physical AI ... (2026-07-19)
- New Horizon daily digest, 2026-07-20 (topic provenance)