HPC News Bytes – 20260720

**Doug Black** (0:05)
Welcome to HPC News Bytes, a weekly show about important news in the world of supercomputing, AI, quantum computing and other advanced technologies. Hi everyone, welcome to HPC News Bytes. I'm Doug Black and with me is Shahin Khan.
Japan plans a giant, nearly 30,000 GPU national AI supercomputer that they're building with NVIDIA. Last week, NVIDIA announced it's partnering with Japan's Notra Corporation to build a 140-megawatt AI factory with 27,500 Rubin GPUs and 13,750 Vera CPUs. The system is called Frontia, F-R-O-N-T-I-A, Japan's state-sponsored physical AI program. The facility will train open multimodal foundation models for robotics, digital twins and industrial automation.

**Shahin Khan** (1:00)
AI is starting to be incorporated into the economies of various countries, so they are trying to keep up with AI by leaning into their strengths. For Japan, that means a bet on manufacturing, robotics, or more generally, physical AI, and using a national platform to pool compute, industrial data, and engineering talent. The US leads in core hardware and software technologies, investment and cloud infrastructure.
China is building domestic chips, open models, and state-backed ecosystems, and Europe is emphasizing trusted AI and industrial applications, while South Korea and Taiwan are reinforcing their semiconductor strengths. At the same time, leadership and frontier models may become less decisive. As AI models tend to commoditize, advantage would shift towards domain expertise, integration, and deployment.
Unless new breakthroughs reset the landscape, like LLMs did, and such breakthroughs could come from anywhere. The sovereignty question is also evolving. In this case, the system is from NVIDIA, but Japan could view it as just a tool, while it retains control over how the system is used. Countries are building AI strengths in different and complementary ways. That may be the really important insight. The theory would be AI is human scale and therefore global in scope, making it so large and complex that realizing its potential and avoiding its pitfalls may require complementary efforts across many countries.

**Doug Black** (2:37)
New export rules for AI chips are coming from the US government. The changes have been expected for some time, but what they might be remains unclear. At a hearing in Washington last week, Undersecretary of Commerce Jeffrey Kessler did not release the Scope Timing Chip Thresholds, Country Coverage or Compliance Rules.
All of this stems from the Trump administration's decision last year to rescind the Biden administration's AI diffusion rule, designed to form a licensing system for AI chips, creating a tiered list of countries and the access they have to US technology. President Trump's Commerce Department officials say the Biden era rule is too restrictive, but to date, they haven't been able to come up with a replacement set of rules.

**Shahin Khan** (3:28)
As we've discussed here a few times, export controls have an important impact, but they also carry costs. Those costs can accumulate even as the benefits dissipate. So controls are most justified when their short-term impact is game-changing, rather than simply adding friction to the existing situation. And that may be the case here. That moves the analysis to the longer term. Broadly speaking, controls can buy time, especially if they meaningfully change the game.
But they also raise the payoff for domestic substitutes, incentivize efforts on architectural efficiency and alternative supply chains, and trigger efforts by the other side to change the game again.

**Doug Black** (4:11)
Staying on the topic of China, it used the World Artificial Intelligence Conference in Shanghai to present a more explicit alternative to the US-led AI order.
Chinese President Xi Jinping promoted open source AI, wider access for developing countries, and a new World AI Cooperation Organization, initially backed by 29 countries. China is positioning the organization around development, sovereignty, and access to AI capability, rather than what they say is the Western emphasis on political values, rights, and safety governance. It is also offering practical incentives, training, cooperation centers, models, and technology that countries can deploy without joining a tightly controlled US ecosystem. This goes beyond diplomatic language, offer models, a technology supply chain, and enough implementation capacity to package AI infrastructure for countries that cannot build it themselves.

**Shahin Khan** (5:12)
Well, speaking of changing the game, China is trying to move the discussion from who has the best technology to first formulating a need for a global AI ecosystem and then who can offer the best possible path towards it. The US provides the leading edge in technologies, know-how, and investment. So China is competing through open models, lower cost infrastructure, financing, and promoting its technology accomplishments, like it did with its CPU only number one system on the top 500 Its narrative also seems to include, shall we say, a different alignment model at a time when US policy appears to demand sharper alignment from partners.