Japan funds domestic foundation model and AI compute
Japan's industry ministry METI and innovation agency NEDO selected SoftBank-backed Noetra and the national research lab AIST to build a domestic multimodal foundation model for physical AI that controls robots, with a fiscal-2026 budget of JPY387.3 billion that could reach up to JPY1 trillion over the five-year FY2026-2030 project if annual milestone reviews are cleared. Separately, METI approved up to JPY72.5 billion in subsidies to five companies, including Sakura Internet and KDDI, to expand domestic GPU cloud capacity under the Economic Security Promotion Act. Named early partners include NEC, Honda, Sony, and major Japanese banks and steelmakers. For AI practitioners, pairing model funding with local compute subsidies could shift cloud vendor choices and data-governance options for robotics and industrial workloads in Japan.
Japan's approach pairs two policy levers that matter for practitioners: direct funding for a named foundation-model developer, and separate subsidies for domestic GPU cloud capacity. That combination is designed to reduce two common bottlenecks for national AI ecosystems at once, model access and affordable local compute, echoing public-private AI strategies underway elsewhere. But the funding structure is worth reading closely: only the first two years are contractually guaranteed, with the rest subject to annual government review.
What happened
Japan's Ministry of Economy, Trade and Industry (METI) and innovation agency NEDO selected Noetra, a company backed by SoftBank Corp., together with the National Institute of Advanced Industrial Science and Technology (AIST), to run a "physical AI" project from fiscal 2026 through fiscal 2030. The goal is a multimodal foundation model combining language, images, video, and sensor data so robots can interpret and act on their environment. METI's fiscal 2026 budget sets aside JPY387.3 billion for the project, funded through GX Economy Transition Bonds, with officials citing total potential funding of up to JPY1 trillion over the five years. Engineers from SoftBank and Japanese AI startup Preferred Networks are expected to join the project, and AIST will coordinate with domestic and international research institutions. Noetra and AIST aim to release an initial foundation model as early as this fiscal year, with annual improved versions after that.
Financial context
The contracts backing the project cover only the first two years; continued funding after that depends on an annual "stage-gate" review that lets the government halt funding if milestones are missed, turning the headline JPY1 trillion figure into a series of pass-fail checkpoints rather than a guaranteed commitment. Named early partners and adopters include NEC, Honda, Sony, Mitsubishi UFJ Financial Group, Sumitomo Mitsui Financial Group, Mizuho Financial Group, Nippon Steel, and Kobe Steel.
Industry context
Separately, the OECD.AI policy navigator documents a complementary METI initiative that approved subsidies totaling up to JPY72.5 billion to five projects to expand domestic GPU cloud capacity under the Economic Security Promotion Act: GMO Internet Group (about JPY1.93 billion), Sakura Internet (about JPY50.1 billion), a joint RUTILEA/AI Fukushima application (about JPY2.56 billion), KDDI Corporation (about JPY10.24 billion), and a joint Highreso/Highreso Kagawa application (about JPY7.70 billion). OECD.AI notes the initiative aims to reduce reliance on foreign cloud providers; domestic firms currently account for roughly 30% of Japan's basic cloud services market.
For practitioners
Pairing model funding with local compute subsidies typically lets domestic teams iterate faster on edge and on-premise robotics integrations, since they can keep sensitive manufacturing data inside national borders and avoid cloud egress costs. Because a state-funded platform can become a de facto integration layer for data formats, safety checks, and pilot standards, suppliers that build around Noetra's platform early may see more direct commercial benefit than the broader wave of AI-linked Japanese firms.
What to watch
Whether Noetra and AIST hit the fiscal-2026 milestone for an initial model release; the outcomes of annual stage-gate funding reviews starting after year two; build-outs and service terms from Sakura Internet and KDDI under the GPU subsidy program; and whether additional manufacturers commit the domain-specific data the project needs for its planned annual model updates.
Key Points
- 1METI and NEDO selected SoftBank-backed Noetra and AIST to build a domestic physical-AI foundation model, budgeted at JPY387.3 billion for fiscal 2026.
- 2Continued funding beyond the first two years depends on annual stage-gate milestone reviews, and METI separately approved JPY72.5 billion in GPU cloud subsidies.
- 3Pairing model funding with local compute subsidies could shift cloud vendor choices and data-governance options for robotics teams operating in Japan.
Scoring Rationale
Japan's coordinated JPY387.3 billion (FY2026, scalable to JPY1 trillion over five years, subject to annual stage-gate review) foundation-model commitment plus JPY72.5 billion in GPU subsidies is one of the larger single-country sovereign AI infrastructure commitments of 2026, directly shaping compute availability, vendor choices, and physical-AI R&D access for domestic teams. Score maintained at 7.0; Finimize and OECD.AI sourcing clarifies the named lead entity (Noetra) and the conditional funding structure behind the headline figures.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
