US chip controls turned Japan into Nvidia’s showcase. China is closing the gap.

On July 16, 2026, Nvidia CEO Jensen Huang unveiled a broad alliance with Japanese industry, from Toyota to Mizuho, anchoring Japan’s sovereign AI build‑out. The centrepiece, a national AI factory named Noetra, will deploy 27,500 Nvidia Rubin GPUs by June 2028, backed by ¥387.3 billion in government funding through March 2027.

Yet the same week, China’s LineShine supercomputer reclaimed the top spot on the TOP500 list, running entirely on domestic chips — a sign that export controls are accelerating China’s homegrown silicon rather than stalling it.

Washington intended its chip export controls to slow China’s AI rise. Instead, the restrictions are splitting the world’s computing infrastructure into two competing blocs — and pushing Nvidia deeper into Japan just as China’s homegrown silicon gathers speed.

On July 16, Huang stood in Tokyo with executives from Toyota, Mizuho, and Kawasaki Heavy Industries, tying Nvidia’s entire AI stack to a nation that had just committed billions to building its own national AI factory. Days earlier, China’s TOP500-topping LineShine system, built with domestically designed accelerators, suggested that export controls are accelerating rather than halting domestic chip development. The summit, billed as a showcase of Japan’s AI ambitions, came three days after Nvidia imposed a stricter compliance whitelist on Asian customers, disqualifying more than half of previous buyers to stop chips reaching China.

The real leverage is quiet: as governments write cheques for sovereign AI, they decide whose full‑stack hardware becomes the default for domestic robotics and industrial AI. Huang even acknowledged a historical debt — a roughly $5 million investment from SEGA in the late 1990s that helped Nvidia survive.

A national AI factory and a web of pilots

At the centre of Japan’s push is the Noetra facility, scheduled to come online in June 2028. It will pack roughly 27,500 Nvidia Rubin GPUs and 13,750 Vera CPUs into about 382 NVL72 racks, drawing an estimated 140 megawatts — enough to place it among the world’s largest AI compute sites. The government has allocated ¥387.3 billion through March 2027, with a long-term target of funneling more than ¥370 trillion into AI, chips, and data centres by 2040. Japan also aims to capture over 30% of the global AI robotics market, projected at about $133 billion.

If Noetra crystallises around Nvidia’s full stack, the chipmaker locks in a multi‑decade customer that other allied governments could copy — a hedge against the permanent loss of China.

That pivot accelerated after Washington extended its chip controls to Nvidia’s newest Blackwell line in May 2026, barring sales to China and forcing the company to deepen ties with allies like Japan. The result is a policy divide captured below.

Beyond Noetra, the partnership web is dense. Mizuho Financial Group plans what it calls the largest on‑premises AI factory in Japanese banking, using Nvidia DGX B200 systems to keep sensitive data in‑house. Rakuten Bank is building transaction foundation models from tens of millions of accounts. Toyota will expand use of Nvidia DRIVE for driver‑assistance and the Omniverse platform for factory simulations. In healthcare, the Tokyo‑1 drug‑discovery consortium — led by Xeureka and expanded in April 2026 to include Astellas, Daiichi Sankyo, Ono Pharmaceutical, and Eisai — runs molecular screening on Nvidia’s BioNeMo platform. Kawasaki Heavy Industries is prototyping surgical, nursing, and transport robots, while Canon and Fujifilm have begun shipping next‑generation CT scanners built on Nvidia GPUs.

Yet most of these are pilots. Concrete deployments include Mizuho’s factory and the CT scanners; the robot prototypes and drug‑discovery benchmarks are early‑stage, and vendors report efficiency gains such as a 15% reduction in rail maintenance costs, though these remain unaudited.

Nvidia’s July 14 compliance whitelist barred more than half of Asian customers, with heightened scrutiny in Singapore and Malaysia — nodes that regulators feared could be used to divert chips to China. Meanwhile, China’s LineShine supercomputer’s top rank, achieved with indigenous chips, shows that export controls are stimulating domestic design rather than suppressing it.

How US controls forged a two‑bloc AI landscape

The separation was not inevitable. Until 2022, Nvidia shipped its most advanced GPUs freely to China. But successive rounds of US export controls, peaking with the May 2026 Blackwell ban, have fragmented the market. Nvidia’s strategy now emphasizes allied governments, with Japan as the most visible showcase. That leaves the rest of Asia — Singapore, Malaysia, South Korea — caught between being AI data‑centre hubs and enforcement nodes.

The upshot is a dual race. Nvidia remains dominant in data‑centre accelerators globally, with AMD and Intel racing to catch up. In China, however, sanctions are fuelling a parallel ecosystem: Huawei’s Ascend chips and Biren Technology’s accelerators power systems like LineShine. Japan’s Noetra and robotics projects tie Tokyo’s industrial future to Nvidia’s stack, while China’s homegrown hardware proves that restrictions can accelerate domestic breakthroughs, not just contain them.

Japan’s regulatory framework is less prescriptive than the EU’s AI Act, leaning on infrastructure incentives and guidelines while leaving gaps on algorithmic accountability. But the FRONTia‑Noetra model makes AI policy an arm of industrial strategy, a choice that other middle powers will study. The US hoped to choke off China’s access. Instead, it has backstopped Nvidia’s reliance on allied government contracts while China races to close the hardware gap. Japan’s July 16 summit is a snapshot of a split computing world.

Beyond the headline

The Power Behind It

The real leverage in Nvidia’s Japan push lies with governments deciding who owns national compute. By writing multi‑year cheques for Noetra and similar sovereign AI projects, Tokyo effectively chooses which vendor’s stack becomes a default for domestic robotics and industrial AI. That, more than any single partnership announcement, determines whether Nvidia or a rival underpins the next generation of Japan’s manufacturing and scientific infrastructure.

The Bigger Picture

Nvidia’s full‑stack courtship of Japan is one manifestation of a broader shift: states treating AI compute like energy or railways, a strategic utility rather than a pure market good. Export controls that shut Nvidia out of China have accelerated this trend, pushing allied governments to build their own AI factories while Beijing doubles down on indigenous chips. The result is an AI landscape increasingly organised along geopolitical blocs, not just corporate competition.

The Reach

Japan’s sovereign AI build‑out, anchored on Nvidia hardware, reaches into Western corporate strategy by reshaping where leading models are trained and deployed. As national AI factories and robotics‑centric ecosystems grow outside US hyperscaler campuses, Western firms in sectors from automotive to industrial automation must decide whether to align with these government‑backed stacks or maintain purely cloud‑centric approaches, altering procurement, standards and collaboration patterns across the global tech industry.

A split supply chain, four immediate decisions

With export controls partitioning the AI chip market and Japan betting its industrial future on Nvidia’s stack, the choices facing Western businesses are no longer theoretical.

• Western semiconductor procurement manager

  You must re‑evaluate supply chains for AI accelerators. Monitor BIS updates on the export‑control whitelist, audit any data‑centre partners in Singapore and Malaysia for compliance, and prepare for potential allocation shifts as Nvidia prioritises allied government projects. A glance at the Federal Register and Nvidia’s quarterly country‑revenue splits becomes essential reading.

• US-based investor with APAC emerging market exposure

  Japan’s AI factory and robotics targets make industrial automation and AI infrastructure a distinct theme. Track Nvidia’s revenue breakdown by country to assess Japan’s share; watch Mizuho’s AI factory rollout as a bellwether for enterprise adoption. For China, the LineShine milestone suggests domestic chip makers are investable, but verify independent benchmarks before acting — vendor‑reported efficiency data is no substitute.

• European industrial robotics manufacturer

  Evaluate whether your next‑generation automation stack should be interoperable with Nvidia’s Omniverse and Isaac platforms. Consider forming a joint research lab with Japanese partners to access the Noetra compute resource. Monitor METI’s FRONTia progress reports; if Japan sustains its 30% global robotics market‑share target, the window to partner or compete narrows.

• Western AI policy professional

  Japan’s FRONTia‑Noetra model offers a middle path between US export controls and EU regulation. Examine how Tokyo’s approach ties AI investment to industrial strategy, and note the gaps on algorithmic accountability. The upcoming METI funding decisions will signal whether sovereign AI becomes a pillar of Japan’s economic security doctrine — a template for nations weighing compute sovereignty.

Explainer

Sovereign AI

A nation’s strategy to build and operate its own AI infrastructure — chips, data centres and models — inside its borders. Japan’s Noetra project is the largest current example among US allies. India recently earmarked ₹10,000 crore for a comparable sovereign AI compute facility.

Full‑stack AI

A vendor’s offering that spans the entire AI pipeline: hardware, servers, networking, software frameworks and pretrained models. Nvidia’s full‑stack includes CUDA, Omniverse, and the Nemotron model family. The approach gained steam after Nvidia acquired networking specialist Mellanox and began bundling its own SuperPOD designs.

Noetra

Japan’s state‑backed national AI supercomputer, set to come online in June 2028 with Nvidia Rubin GPUs. Noetra aims for a power usage effectiveness (PUE) of around 1.2, markedly lower than the global data‑centre average of 1.6. That efficiency target is part of METI’s plan to make the facility a model for green AI compute.

FRONTia

Japan’s METI‑led program to fund national AI research infrastructure, most prominently the Noetra facility. FRONTia channels public money into open multimodal foundation models intended for manufacturing, logistics and healthcare. The project is part of a broader economic security push to reduce reliance on foreign hyperscalers.

TOP500

A biannual ranking of the world’s most powerful supercomputers, maintained since 1993. Systems are measured using the LINPACK benchmark, which solves dense linear equations. China’s LineShine topped the June 2026 list with fully domestic chips, a first for a Chinese system running exclusively on indigenously designed processors.

Genesis Mission

A US government initiative to apply AI to scientific grand challenges, covering climate, health and energy. Japan was named the first international partner during Huang’s July 2026 summit. The mission links Japan’s research base to the US Department of Energy’s national lab network and its AI‑for‑science platforms.