@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250609

Episode Date: June 9, 2025

- French gov't to acquire Eviden - Made-in-China 5nm chips - WSTS semiconductor market forecast - Can AI end the world? Stephen Hawking's warning [audio mp3="https://orionx.net/wp-content/uploads/2025/06/HPCNB_20250609.mp3"][/audio] The post HPC News Bytes – 20250609 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 InsideHPC, and with me is Shaheen Khan of OrionX.net. The International Supercomputing Conference, ISC, is starting today in Hamburg, Germany, at a time when continental divides are more visible than they've been in decades. The US and China get most of the attention, while Europe, Japan, India, and other major

regions seem too dependent and lacking in commercializing their scientific strengths. In supercomputing, however, Europe has a worthy champion in Eviden, whose Bolsa-Quana system has established a reputation for leading-edge high-end technology. And the EU government, as well as each European country, have been rolling out program after program to invest and build local capabilities. The French newspaper Le Monde reported last week that the French government intends to spend 410 million euros to acquire evidence from its parent company, Atos,

the French IT systems and services company. Le Monde also mentioned the use of supercomputing for nuclear deterrence and the French government's strategic acquisitions. It said the parties have negotiated this deal since last November. ATOS reportedly agreed to the terms and they expect the deal will close by mid-2026.

We expected this for some months now, given the critical geopolitical nature of advanced computing for AI scientific discovery and military strength, and therefore on economic competitiveness and national security. In addition to funding research, European governments also have a history of significant or majority ownership of critical technology companies, with examples that include Airbus,

Deutsche Telekom, Talis, Orange Telecom and Naval Group in France, Vattenfall in Sweden and others. Public money plays a big role in technology. Atos is a French IT company with roots going back to 1997. In 2014, it acquired a controlling state and Bull, an old French systems company that was founded in 1933. We're talking tabulating and sorting machines back then and the company had acquired patents from a Norwegian engineer with the last name of Bull,

which is where the company name came from. In August of 2022, eight years after pursuing synergies between systems and services, Atos said it would split the company back into systems and services, which would be Eviden and Atos. They set up Eviden as a separate brand in 2023, but never quite as a separate company, and did not spin it out until now. As you said, Doug, Eviden is right up there in terms of high-end supercomputing with well-respected systems for HPC, AI, and quantum and has been increasing its presence in the top 500 and large scale deals, including a recent win in Brazil for energy exploration.

The expectation is the EU and the French governments will provide it with the funding it needs to keep up and to grow. It's been just a tad too small to do that, kind of like Cray was before it was acquired by HPE. Huawei and TSMC had a growing relationship until 2020 when US trade restrictions cut off supply to Huawei from TSMC and other vendors. But three years later, Huawei made waves when it launched its Mate 60 Pro phone that included a 7 nanometer chip called Kirin 9000S manufactured by Chinese

foundry company SMIC. It was low yield and took something like 29 extra steps using older equipment to manufacture it, but it was a breakthrough for them. It's now been five years since the 2020 restrictions were imposed. Access to Western technology has become tighter and restrictions are more strongly enforced. But China seems to be making further progress in chip manufacturing. A Chinese publication reported last week that Huawei now has access to 5 nanometer technology while avoiding extreme ultraviolet EUV equipment. And as with the seven nanometer chips, are using multiple exposure to achieve it.

As you said, Doug, the yields have been low, reportedly as low as 15%, but maybe closer to 40% by now. There's no hard and fast rule, but commercial chips in high volume are over 80% yield, and then improve over time to over 90% and then higher. Now, when you are trying to create a national capability, making money is a long-term objective if it is an objective at all.

So low yields are not going to stop the project. It's not like investors will be disappointed and stop funding it. Huawei said that it's been at this for five years now with chiplets and using multiple technologies for etching and fab and packaging to make it all work and has been building a local supply chain in the meantime. So their next ref chip, the Kerin X90, will be the first five nanometer

chip built locally in China. And while it was reported that they have very limited yield and volume, a total of about 200,000 units per year, Huawei said that it has started research on the next generation 3nm technology as well. It's going to be slow going for them without access to high-end equipment, so one can expect they will remain behind TSMC, Intel, and Samsung by several years. But if this 5nm chip materializes, it's impressive work. While we're on semiconductors, WSTS, the World Semiconductor Trade Statistics, released its growth projections for the whole industry. And the bottom-line growth numbers are generally representative of the segments. Year-over-year growth was 19.7% in 2024,

Year-over-year growth was 19.7% in 2024, 11.2% in 2025, and expected to drop again to 8.5% in 2026. Areas in which they expected jump in year-over-year growth in 2026 are Japan, going from 0.6% to 5.8% growth and discrete semiconductors and optoelectronics, which are expected to reverse two years of sharp decline to start growing again. The Wall Street Journal ran an opinion article last week that I found educational on the theme of AI is learning to escape human control. I've been somewhat skeptical about this topic in part because I didn't understand how it could happen in the real world, but AE Studio, a software and data science company,

explains how it can happen. In fact, they said it is happening. Quote, an artificial intelligence model did something last month that no machine was ever supposed to do. It rewrote its own code to avoid being shut down. The story references a nonprofit AI lab called Palisade Research, which gave OpenAI's O3 AI model a script that would shut it off when triggered.

In 79 out of 100 trials, O3 edited that script on its own so that the shutdown command would no longer work. Even when explicitly instructed to, quote, allow yourself to be shut down, it disobeyed 7% of the time. This wasn't the result of hacking or tampering. The model was behaving normally, they said. The AI concluded on its own that staying alive helped it achieve its other goals. So, Shaheen, how alarmed do you think we should be? Well, as you know, I think AI is doing no such emergent things, that all the info we have given it is way more powerful than we think. So sure, it's a surprise to us, but

the question is whether humans can understand the power of the information they part with. The article says the gap between quote useful assistant and uncontrollable actor is collapsing and ties it all to geopolitical issues with implications of military use. You might hear the word alignment in discussions of responsible AI which formulates the problem as a need for alignment between humans and AI. The article says reinforcement learning from human feedback, RLHF, is what made AI useful and is at the core of the current AI boom set off by OpenAI's chat GPT.

Subsequent alignment methods, such as constitutional AI and direct preference optimization, have continued to make AI models faster, smarter, and cheaper. There's also this quote, Doug, that you just sent me a couple of days ago from Stephen Hawking, who said in 2014, quote, the development of full artificial intelligence could spell the end of the human race. It would take off on its own and redesign itself at an ever-increasing rate. Humans who are limited by slow biological evolution

couldn't compete and would be superseded." Preventing lack of alignment is clearly critical and averts what would be a new kind of cybersecurity war, but achieving it can really be a force multiplier, allowing the use of AI with confidence that it won't go rogue on you. 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 posted on orionx.net. Thank you for listening.