@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250505

Episode Date: May 5, 2025

- Japan's Rapidus 2nm chips - McKinsey's $7T datacenter forecast - Nvidia, trade restrictions, national competitiveness - Geoffrey Hinton's AI warning [audio mp3="https://orionx.net/wp-content/uploads/2025/05/HPCNB_20250505.mp3"][/audio] The post HPC News Bytes – 20250505 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC News Bites. I'm Doug Black of Inside HPC, and with me is Shaheen Khan of OrionX.net. Last March, we covered Japan's efforts to reassert itself in the chipmaking arena. In the late 1980s, the country had more than 50% of the global chip market. That was also when Japan's fifth-generation computer systems project pursued, with some

fanfare, AI systems via logic-based programming and parallel processing. It increased trade tensions with the U.S. that had been on and off with textiles and steel and had intensified with cars and then expanded into chips. It also started before big shifts in technology like the internet, GPUs, and extreme ultraviolet EUV lithography. In 2024, it is estimated that Japan's market share in chip manufacturing equipment is about 30%, while its share in chip manufacturing is about 10%. AI systems will proliferate into devices,

but for now, the AI market is mostly in the data center, and Japan's share of data center chip manufacturing is reported as 3.3%. So it's been quite a dramatic drop, but that could be about to change. Wouldn't you say, Shaheen? Well, the 2 nanometer chip manufacturing arena has been limited to TSMC and Intel with expectations that Samsung would join. But now there's a new player that's got a lot going for it.

It is a company called Rapidus in Japan, which seems on track to go into 2nm production in 2027. It said its pilot production line started on April 1st, 2025, and will be in full production by end of April. Rapidus was formed in 2022 by a who's who consortium of Japanese companies with a five year mission to produce two nanometer chips in Japan. In alphabetical order, the consortium

is comprised of Denso, Kioxia, MUFG Bank, NEC, NTT, SoftBank, Sony, and Toyota. Rapidus also has strategic collaborations with IBM, whose chip research division was first to demonstrate a 2 nanometer node using its 300 millimeter wafer. As you'd expect, Rapidus has been nicely funded. $6.2 billion so far from the Japanese government, with another $5.4 billion pledged, securing equipment from ASML, which is forming a local facility with some 50 engineers to support rapidus, and the IBM collaboration on node and packaging technologies.

But it needs a lot more money, tens of billions of dollars more, that it looks like it will raise privately with the help of all of the above. NVIDIA CEO Jensen Wong made a splashy appearance at a conference in Washington last week, saying, among other things, that China is not far behind the US in AI and that Chinese tech giant Huawei is developing powerful AI compute capabilities. Shaheen, I didn't hear all of Jensen's remarks, but I'm not quite following his logic when he argues against trade restrictions on some Nvidia GPUs to China, specifically VH20, which is a less powerful AI chip than Nvidia's top-of-the-line processors. According to a CNBC article,

focus on making its companies competitive and that restricting chip sales to China and other countries threatens US technology leadership. But from what we've heard from analysts with intimate knowledge of the scene in China, such as our podcast guest, Dr. Handel Jones, chip export controls have significantly held back China's progress in AI. And even if the H20 was barred from exports to China, that wouldn't stop Huawei from developing their own AI chips, correct? Yeah, the reporter's question seemed to be based on a formulation where the US is way ahead in AI. And the reply was a recalibration of that view, really implying that certain options and objectives may not be available. For example, just to make it up, if China has the equivalent technology for something

within its borders, then stopping the flow of that something does not hurt them, but it does hurt the US vendor of that something and also limits exposure to the market there, things like that. As we have said here, Nvidia is becoming more visible in its commentary on global trade while carefully recognizing the importance of chips and AI on national security and the government's important role to set policy. It also comes across as recognizing the complexity of the situation, aiming to clarify the dynamics rather than sound critical.

You may recall I was the chief competitive officer at Sun Microsystems for a while. And in any competitive situation, it is critically important to have a very accurate assessment of your competition's so-called SWOT, strengths, weaknesses, opportunities, and threats. Overestimating or underestimating them both lead to mistakes. If you are way ahead, you have certain options. And if you are neck and neck, you have certain other options. Reality must drive the strategy, while the tactics can be over the map,

as long as they don't negatively change the reality. It is one reason geopolitics and national competitiveness is 3D chess, and why the situation between the US and China has been described as much more complicated than how it was in the Cold War with Russia. A new report from McKinsey says that by 2030, data centers globally will need to invest $6.7 trillion to keep up with the demand for compute power.

Quote, data centers equipped to handle AI processing loads are projected to require $5.2 trillion in capital expenditures, while those powering traditional IT applications are projected to require $1.5 trillion in capital expenditures. Overall, that's nearly $7 trillion in capital outlays needed by 2030. It's all fundamentally HPC workloads, but it's also a big contrast with scientific research funding. Only a few short years ago, the $500 million Exascale project was viewed as huge and maybe in the realm of hyperscalers, but certainly not private industry as a whole. But now we see a stampede

to large scale computing everywhere. At the same time, funding for traditional HBC has not kept pace. And the cuts that we are hearing about at the NSF and NIH and other federal agencies will amplify that, sure to influence the size of the traditional HPC market. There was news, for example, that the 400 petaflop, $457 million Horizon supercomputer project at Texas Advanced Computing Center, TAC, at the University of Texas,

really a flagship center funded by the National Science Foundation is now at risk. It's already spent $100 million on the system, and we'll watch and see what happens. We think it's probably a good idea now and then to report on those who raise red flags about the potential risks of AI. We mostly talk about companies pushing AI as far and as fast as possible, and it's really not in their interest to warn us about AI's possible dangers. Even when they have issued such warnings, such as Elon Musk

in 2017, his companies have pursued AI as aggressively as they can. So last week, no less than, the godfather of AI, Nobel laureate Jeffrey Hinton, warned there's a 10 to 20% probability that AI will come to take control over humanity. Hinton is a pioneer of AI based on neural networks, the fundamental approach that ended the AI winter when it was combined with lots of data and powerful hardware. And that led to a similar pattern with generative AI and now agentic AI.

But Shaheen, looking at the state of the world these days, let me pose the idea. Maybe it wouldn't be all bad if AI actually does take over. What say you? Well, Doug, the idea of replacing politicians with AI that works may just change the public sentiment about AI and gain new global support for Silicon Valley. But anyway, when Hinton speaks, one must listen.

He famously left Google in 2023, not as a protest since he seemed good and even impressed with how Google was pursuing responsible AI, but to be able to openly discuss his concerns about the dangers of AI. We certainly know about weaponized information aided by AI to manipulate societies, social disruption as new jobs are not accessible

to those whose job is eliminated by AI, and the possibility, some people think inevitable certainty, that AI becomes so much smarter than humans that it will look at humans like humans look at pets. AI red flags have been there from the beginning when robots were the focus. After all, Isaac Asimov's original Three Laws of Robotics was published in 1942. It continues to cover it well. Over at OrionX, we referenced it in our major paper on AI about 10 years ago. The laws, just to refresh your memory, say one, a robot must not harm a human being or through inaction allow a human being to come to harm. Asimov changed this to humanity or mankind in 1985. 2. A robot must obey orders given it by human beings except where such orders would conflict with the first law.

And 3. A robot must protect its own existence as long as such protection does not conflict with the first or second law. Quite nice. All right, that's it for this episode. Thank you all for being with us. HPC News Bytes is a production of OrionX in association with Inside HPC. Shaheen Khan and Doug Black host the show. Every episode is featured on insidehpc.com and and posted on OrionX.net. Thank you for listening.