Nvidia’s Japan blitz: Fujitsu, FANUC, Toyota partnerships, Jetson T3000, and the Vera Rubin strategy

Japan partnerships announced in Tokyo

Fujitsu-led business study

Fujitsu announced it is exploring the development of a "collaborative physical AI platform" integrating Nvidia technology, with FANUC, Yaskawa Electric, and Kawasaki Heavy Industries as initial partners . The initiative is not a single joint venture: each participant is at a different stage of commitment. Fujitsu is running a business study; the three robotics firms are integrating Nvidia technologies under that study; and all four — along with 18 other Japanese organizations — intend to join the Nvidia Cosmos Coalition to co-develop open physical AI models . The goal, as stated by Fujitsu, is “promoting the development of a collaborative control platform that ensures sovereignty by bridging the digital and physical worlds” .

Cosmos Coalition

22 Japanese organizations — including AIRoA, FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, Kubota, NEC, SoftBank Corp., Sony Group Corporation, and Yaskawa Electric — intend to join the Nvidia Cosmos Coalition . The coalition is Nvidia’s mechanism for collaborative development of "open frontier physical AI models," where partners jointly build and share models rather than keeping them proprietary .

Toyota collaboration

Nvidia separately announced expanded collaboration with Toyota Motor Corp. Nvidia will provide advanced GPUs to Noetra Corp., a new company established by Toyota along with SoftBank, Sony, Honda, and NEC to develop domestic AI technology . Noetra began operations on July 16, 2026 .

What did not happen

Despite the blockbuster optics, there was no single joint venture, no disclosed shared investment amount, no common rollout date, and no unified ownership structure binding all the companies together . The announcements are better understood as a coordinated but decentralized set of commitments: some participants are studying a platform, some intend to join a technical coalition, and others are pursuing independent product development using Nvidia’s robotics and vision tools .

New edge AI hardware: Jetson T3000 and T2000

On July 15–16, 2026, Nvidia introduced two new compact modules based on the Nvidia Thor architecture, designed to run foundation models directly on autonomous machines, humanoids, and general-purpose robots .

Both modules are based on the same Thor SoC that powers the existing Jetson Thor lineup, but are significantly pared down in configuration and power . An industrial variant, IGX Thor T3000, adds functional safety features .

Availability: Both modules are scheduled for launch in the first quarter of 2027 — they are not shipping this year . Developer emulation for the T3000 is available starting July 2026 via JetPack 7.2.1 .

The broader strategy: Open models + Vera Rubin

These announcements are not isolated. They form a coherent three-layer strategy spanning open models, edge hardware, and datacenter-scale AI compute.

Layer 1: Open model layer — Cosmos 3 Edge

Alongside the hardware, Nvidia unveiled Cosmos 3 Edge, a downsized version of its physical AI foundation model optimized specifically to run on the new Jetson Thor modules . This is the software counterpart to the hardware announcements: a lightweight model (4 billion parameters) designed for on-device real-time inference and action generation .

Layer 2: Edge hardware layer — Jetson Thor (T3000/T2000)

The new modules provide "right-sized" compute to run those open models directly on factory robots, eliminating the need to connect to the cloud for inference . This is the mass-market "right-sizing" of robot compute that Nvidia has been driving toward — balancing performance, power, and cost for widespread deployment rather than pushing the highest possible specs .

Layer 3: Datacenter/cloud layer — Vera Rubin platform

At the high end, Nvidia’s Vera Rubin platform — its next-generation AI supercomputing architecture — entered full production earlier in 2026. At CES 2026, Jensen Huang confirmed Vera Rubin is in full production, with partner systems shipping in the second half of 2026 . Nvidia’s official newsroom stated Rubin-based products will be available from partners in H2 2026 . By May 31, 2026, Nvidia announced Vera Rubin was ramping into full production at scale with Taiwan’s top server makers . The platform uses a Rubin GPU with up to 288 GB of HBM4 memory and the Vera CPU .

The strategic thread

Japanese factories equipped with Cosmos 3 Edge open models running on Jetson Thor hardware at the edge — all trained and orchestrated by Vera Rubin supercomputers in the cloud or at AI factories. Nvidia is simultaneously building the cloud-scale AI infrastructure (Vera Rubin, shipping H2 2026) and the on-device edge AI hardware (Jetson T3000/T2000, shipping Q1 2027) to make physical AI in manufacturing a reality, with Japan’s industrial giants as the launch customers .

As Nvidia’s own blog put it after Huang’s lunch with the CEOs of Fujitsu, Kawasaki, FANUC, and Yaskawa: “These are the companies that will move physical AI from national ambition into the manufacturing floor” .