@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260309

Episode Date: March 9, 2026

- AI+HW 2035: Shaping the Next Decade - CPUs are in short supply too - What is causing IT shortages - Foxconn is upbeat about 2026 - Foxconn, its rivals, and its customers - EuroHPC Summit 2026 postponed [audio mp3="https://orionx.net/wp-content/uploads/2026/03/HPCNB_20260309.mp3"][/audio] The post HPC News Bytes – 20260309 appeared first on OrionX.net.

Welcome to HPC Newsbytes, a weekly show about important news in the world of supercomputing, AI, quantum computing, and other advanced technologies. Hi, everyone. Welcome to HBC Newsbytes. I'm Doug Black, and with me, of course, is Shaheen Khan. More than two dozen researchers have collaborated on a paper published on the archive site called AI Plus HW 2035, shaping the next decade. They call for the global research community to pull together, a, quote, cohesive, long-term vision to strategically coordinate the development of AI and hardware, unquote, to address fragmentation that is constraining progress toward holistic, sustainable, and adaptive AI systems

capable of learning, reasoning, and operating efficiently across cloud edge and physical environments. This is a visionary and certainly well-intentioned, along with well-thought-out, call to action, but one that could amount to an exercise in cat herding. Question, what are the key catalysts that make a broad-based effort like this succeed? Certainly, leadership from a leading company that has embraced the spirit of the project, and that, of course, means invidia. And in fact, Nvidia is represented among the paper's authors. Or if Nvidia decides not to take part,

then a company like AMD setting up an alternative AI hardware software ecosystem in competition with the market leader. Shaheen, the paper's abstract seems to hit the key issues, but as you read it, were there problems or challenges you think they left out? Well, I went back to the paper that we covered two episodes ago by Jack Dungera, Dan Reed, and Dennis Cannon, that was titled Ride the Wave, Build a Future, Scientific Computing in an AI world,

which was basically pursuing a similar thought process objective. and it looked at the impact of AI on systems from specifically an HPC perspective. The focus of this new paper is not specifically HPC, but there is quite a bit of overlap between the two papers, which reflects the overlap between HPC and general computing as well. So, for example, energy efficiency, taking center stage and new emphasis on energy metrics, like performance per watt or performance per jewel as the other paper would have,

Co-design principles that came out of HPC, which the new paper mentions together with holistic cross-layer optimization for hardware algorithms and applications. That's how it formulates it. Hybrid computing to support integrated workflows that can combine AI simulation, physics-based AI analytics, and edge computing, which also points to devices and instruments and any and all connected physical systems and data centers of various. sizes along the way to cloud, that whole expanse of things, all integrated into a single logical computer that spans from sensors to supercomputers and everything in between. And finally, national level scale, which both papers recognize to be important and requiring public-private

partnership. The new paper is really a hardware research agenda, as its title indicates, AI plus hardware 2035. And it mentions familiar topics in the new paper. in broad brush terms in this rendition of the paper. And those are things like memory-centric architectures, 3D integration and advanced packaging, domain-specific accelerators, specialization, customization. We've talked about that a lot in this podcast. And then using AI to assist in design, so-called AI in the loop. And, of course, targeting a thousand times improvement in AI efficiency over the next decade.

Sounds good to me. Now, with the acute shortage of parts, memory prices jumping every week, and a new war that's impacting energy supplies, you might think the IT world is bracing for the worst. But not if you listen to Foxcon and AMD. AMD CEO Lisa Sue was reported last week saying demand for its CPUs will likely exceed supplies thanks to, or could be blamed on, rapidly growing AI inference workloads. Just over a month ago, both Intel and AMD reportedly notified their big customers that CPU supplies will be tight. Delivery lead times were said to be up to six months for some Intel CPUs and up to 10 weeks for some AMD products. Well, there is a shortage of everything now, really, from raw materials to memory wafers, network switches, obviously GPUs, and even hard disk drives. And now to CPUs, which have been kind of taken for granted in recent years.

Everything can run on CPUs and many things really need CPUs. And CPUs have onboard vector and matrix extensions that make them more capable, but they can't compete with GPUs for heavy number crunching. So for AI inference and agentic AI, the preferred configuration for a system continues to include GPUs in abundance. But AI agents put more load on CPUs for quote,

planning and action, end quote. Things like handling logic, and managing library API calls, and many sequential tasks that are ideal for CPUs. At the same time, smaller AI models have become better, and in addition, many agentic AI applications can afford to wait a bit. So a CPU-only or a CPU-heavy system would be simpler and more cost-effective. But why do we have these shortages? The short answer is a zero-sum manufacturing situation that has materialized for memory and logic semiconductors. For logic, shared high-end manufacturing capacity creates a zero-sum situation for CPUs and GPUs. That capacity is itself in short supply,

or already spoken for. So if capacity can be redirected to another product at all, it would likely go to GPUs, which would be more profitable. Of course, it still cannot meet that demand, but in the process, it creates a shortage for CPUs as well. It's a similar situation when it comes to memory. Memory wafers are in short supply, and the same wafer can be used for a variety of products, from high bandwidth memory that is used on Nvidia and AMD GPUs, to solid-state disks and DRAM.

So wafers were redirected to the more profitable HBM. They still cannot meet HBM demand, but in the process cause a shortage for other related products. as well. So big ripple effects, which complicates planning and procurement processes for customers and the price quotes, revenue management, and catalog price update processes for vendors. All that complexity does not seem to be a factor in the short term, at least for the big players. It's been reported that Foxconn, the big Taiwanese contract manufacturer, is upbeat, saying the Middle East War and memory shortages and price increases would have little impact on its

short-term business and expects double-digit revenue growth in 2026 to more than 280 billion. Company said AI, HPC, and a revived consumer electronic sector were the main drivers. Foxcon's projections, as you alluded to, are related to its large size and the large size of its customers, which gives them market clout and enables them to secure capacity and negotiate better prices. Now, Foxconn is a huge company, as you indicated, also known by, its Chinese name, Hanhai Precision Industry, it is the largest contract electronics manufacturer in the world and has been for well over a decade. It's four or five times bigger than its nearest rival and about the size of its top ten competitors combined. Foxcon's rivals include Quanta,

Wistron, Wistron, Supermicro, and Sanmina in the data center space. Pagotron, Luxshire, Tata in the consumer space, and Tata actually acquired Wistron facilities in India to be bigger than it was. was before, and Flex, J-Bell and Sanmina in contract manufacturing. Several of these are also based in Taiwan, making Taiwan the powerhouse that it is not just in chips. Foxcon's customer base is really the Magnificent Seven and other big tech companies. Apple accounts for more than half of Foxcon's revenue, and Foxconn also makes Sony PlayStation, Nintendo's, and consumer devices for Dell.

On the data center side, Foxconn is reported to be the largest manufacturer. of NVIDIA servers. At the GTC conference next week, Foxcon is expected to showcase complete systems based on NVIDA's new VeraRubin chips. It also builds systems for big cloud providers and hyperscalers, like Amazon, Microsoft, Google, Meta, and it's working with OpenAI and also with SoftBank

for the National Stargate AI Initiative in the US. Now, Apple and Foxcon are the main characters in the 2025 book, Apple in China, written by Financial Times reporter Patrick McGee. The book provides a description and a critique of Apple's role in building China's electronics manufacturing capabilities and in return benefiting from that collaboration to be as big as it is now.

We'll close with the news that the EuroHPC Summit organized by the EuroHPC Joint Undertaking and scheduled for this week, March 10th through 12th, in Cyprus, has been postponed until further notice. Their statement said, the EuroHPC Summit, 2026 is postponed due to the evolving situation in the Middle East, which has led to travel disruption and limited flight availability. We are currently exploring possible new dates with the Cyprus EU presidency. Yes, lots of flight cancellations, really across Europe, not just as you get closer to the Middle East, which is really another example of disruption that is rippling through a global supply chain. All right.

That's it for this episode. So thank you all for being with us. HPC Newsbytes is a production of OrionX. Shaheen Khan and Doug Black host the show. Every episode is posted on OrionX.net. If you like the show, please rate and review it. Thank you for listening.