• Jensen Huang’s mid-July visit came with dozens of Japanese partners across healthcare, finance, robotics and quantum computing. 
• The unifying pitch, “sovereign AI“, is also a pointed reminder of the market NVIDIA can no longer sell to.

Jensen Huang arrived in Japan this week, his first visit in about nine months. Along with him came a partner list long enough to read as a map of the country’s economy. Toyota and Kawasaki Heavy Industries in mobility and robotics. Mizuho and SMBC Group’s Japan Research Institute in banking. Canon and Fujifilm in medical imaging. RIKEN, the national research institute, in supercomputing. And SEGA, the games company, by Huang’s own retelling, kept a struggling Nvidia alive with roughly $5 million in the late 1990s.

The showcase, published on Nvidia’s blog and paired with a separate announcement built around its Nemotron open models, is being staged as more than a product refresh. Huang has cast it in national terms; briefings ahead of a government-and-industry partnership set for today (July 16) described it as “the day Japanese AI begins.” Strip away the choreography, though, and the more useful read for anyone tracking Asia’s chip and AI economy is simpler: Japan is now getting the full-stack courtship Nvidia can no longer extend to China.

Sovereign AI as the organising pitch

The throughline across the announcements is “sovereign AI“, the argument, which Huang has repeated throughout the trip, that a country should build and run its own AI rather than rent it from abroad. It is a framing tailored to allied capitals, and Nvidia’s vehicle for it in Japan is its Nemotron family of open-weight models, released with datasets and training recipes that companies can fine-tune on local data and deploy inside their own borders.

Several Japanese groups are already building on that foundation. The Institute of Science Tokyo trained its Swallow foundation models on Nemotron data, while SoftBank subsidiary SB Intuitions used the same libraries for its Sarashina series; Sakana AI is folding Nemotron into its Fugu model-routing platform, and NTT DATA, Hitachi and ENEOS are building Japanese-language applications on top. 

The commercial logic underneath the sovereignty language is straightforward. Open models that can be governed locally still need somewhere to run, and that somewhere, for now, is Nvidia silicon. For Japan, the pitch also lands against real pressure rather than pure marketing. 

Nvidia and its partners repeatedly tie the push to the country’s ageing population and shrinking workforce, framing AI and robotics as a way to sustain productivity as labour thins out. 

Whether GPUs are the answer to demographic decline is a longer argument, but the demand is genuine, and the government’s appetite is real, enough that Huang has publicly wrapped Japan into a “sovereign AI” strategy he says now spans more than 20 countries.

A sweep across the economy

The breadth is the point Nvidia most wants noticed. In healthcare, the Tokyo-1 drug-discovery consortium run by Xeureka, now including Astellas, Daiichi Sankyo, Ono Pharmaceutical, and, since April, Eisai, is running screening and generative-design workloads on Nvidia’s BioNeMo platform. 

Startup SyntheticGestalt unveiled two molecular models it says lead public drug-discovery benchmarks, a claim drawn from its own testing rather than independent evaluation. Kawasaki is developing surgical-support, nursing and transport robots, and both Canon and Fujifilm have begun shipping next-generation CT systems built on Nvidia GPUs.

In finance, Mizuho plans what it expects to be the largest on-premises AI factory in Japan’s banking sector, starting on Nvidia DGX B200 systems, a deliberate on-prem choice for a bank that wants agents developing against sensitive data without it leaving the building. SMBC’s Japan Research Institute has deployed a Nemotron-based AI factory, and Rakuten Bank is building transaction foundation models on the group’s tens of millions of banking and card accounts.

Nvidia’s X post

Physical AI is where the trip’s rhetoric runs hottest. Toyota is extending an existing partnership to build driver-assistance systems on Nvidia’s DRIVE platform, run factory simulations in Omniverse, and, via subsidiary Woven, train a vision-language model for urban traffic. Nvidia’s Metropolis toolkit, meanwhile, is being taken up by OMRON, Fujitsu, Hitachi and Shimizu for factory inspection, rail maintenance and construction-site safety. 

The efficiency figures attached to these deployments, a 15% cut in rail maintenance and energy costs here, weeks-to-days time savings there, come from the vendors and their partners, and are best read as directional rather than audited.

At the research frontier, two RIKEN supercomputers built on Nvidia’s Blackwell chips are coming online, including a quantum-HPC system linked to on-premises quantum processors. Nvidia also positioned Japan as the first international partner in the US government’s Genesis Mission, its AI-for-science initiative, a framing that quietly folds Japan’s research base into a US-led scientific bloc.

The read for the rest of Asia

The part that makes this a regional story, rather than a Tokyo one, is what sits just off-frame. Nvidia has told investors it is effectively foreclosed from China’s data-centre market, after Washington’s export controls tightened again in May to cover its Blackwell line. On July 14, days before the Japan showcase, the company rolled out a stricter compliance regime for Asian customers, reportedly a whitelist that disqualified more than half of past buyers, with added end-user scrutiny in Singapore, Malaysia and Japan aimed at stopping chips from being rerouted into China.

Seen against that, the Japan blitz is the positive image of the same policy. The market Nvidia can sell to freely, it is now saturating: government backing, open models tuned to the local language, and hardware embedded across banks, carmakers and hospitals. The market cannot, it is walling off, including through the Southeast Asian intermediaries that had become a suspected leak point. 

For Malaysia and Singapore, courted as data-centre hubs but now also policed as potential loopholes, that dual role is worth watching closely. There is also a competitive subtext that Huang did not dwell on. China’s LineShine system reclaimed the top spot on the June TOP500 list, a reminder that export controls have accelerated, not stalled, Beijing’sdomestic effort, and part of the reason Nvidia is so keen to lock in allied ecosystems while it still holds a lead.

What none of this settles is how much of the Japan showcase converts from pilot to production. The strongest signals are the on-premises factory commitments from banks and the CT systems already shipping; the softest are the demo-stage robotics and the benchmark claims that arrive pre-packaged from the companies making them. 

The honest through-line for readers is that Nvidia has made Japan the clearest expression yet of what its favoured-market strategy looks like, and that the sovereignty it is selling still runs, for now, on someone else’s chips.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.

Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.