@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20251124

Episode Date: November 24, 2025

- SC25 quick takes - Thank you St. Louis - Hyperion Research annual SC briefing - Hyperion quantum computing update - New public-private partnership model for DOE's new supercomputers [audio mp3="https://orionx.net/wp-content/uploads/2025/11/HPCNB_20251124.mp3"][/audio] The post HPC News Bytes – 20251124 appeared first on OrionX.net.

Welcome to HPC Newsbytes, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC Newsbytes. I'm Doug Black of InsideHPC, and with me is Shaheen of OrionX.net. Shaheen and I are fresh off of SC25, and we do plan to do a more thoroughgoing post-conference look back, but a few top-line impressions, Shaheen would include, that SC has definitely grown compared with the pre-pandemic period. Attendance last year in Atlanta swelled to 18,000 plus, while this year it was about 16,500, but some of that decline can be attributed to the government shutdown, not ending soon enough for many federal employees to get travel plans

approved. And also, there were concerns among Europeans by traveling to the states. That said, 16,500 is well above the typical 10 or 11,000 attending the conference previous to 2020. And also, we should say this year saw a record number of exhibitors. Right. 560 of them, which was a massive increase from last year. In terms of advanced technology, the conference continues to be the best window into the future and an early adopter paradise, if you will. What do you see at SC? You see everything. You see networking. You see visualization. You see user interface. You see robotics. You see all of the above. So a quick sampling could be a new 64-bit Xaflop system on the top 500, which we covered in our podcast last week. Liquid cooling, which

continues to be front and center again. Quantum computing, making steady progress, but not there yet. New CPUs and GPUs and architectural approaches, which are very exciting. Interconnects. PCIE 8 was announced. PCIE 7 specifications are out. You could see PCIE 6 hardware at the booth running demonstrations. Ultra Ethernet Consortium, Ultra Accelerator Link, UALink. Those are all topics that were covered in some detail. Open source and the impact that it has in this industry and on the IT market in general. Mixed precision, which obviously is what drives a lot of the debate about what is HPC, what is AI. In fact, the whole HPC AI quantum, what is what, what is a subset of what, and are they

different segments with debates that you could participate in storage, optical communications, but also optical computing, maybe. There were some examples of that and much, much more. Also, I want to send our thanks to St. Louis, the city. It's a beautiful city with a lot of history, of course, Gateway to the West, with the world famous Gateway Arch, and the legendary and must try. gooey buttercake, a tradition in St. Louis, and we had samplings of that at some of the parties, and they were absolutely awesome. And also, it happens to be the home of the World Chess Hall of Fame,

and they also have a way cool, Godi-esque city museum that is also a must visit. So thanks to St. Louis. And finally, SC is a wonderful reunion. It's a reunion of a community that warmly accepts new members. So it's a great way to reconnect with old friends and make new friends. So looking forward to next year's SC26 in Chicago. Now, an SC tradition, and really it's a social and business must, is the annual breakfast market update put on by Hyperion Research during the conference. The firm's CEO, Earl Joseph, said the HBC AI sector, and by that he means AI driven by HPC class technologies as opposed to social media AI, grew at more than 20 percent the first half of this year, and that for 2024 it grew more than 23 percent compared to the historic 7 to 8

percent normal rate. In short, the combination of HPC and AI is a juggernaut. Joseph said new technologies and new use cases are expanding quickly. AI and quantum computing systems are being installed at a high rate. Governments around the world are making major AI investments, well above anything in the past. AI factories and gigafactories are going up in Europe. New AI sites in the U.S. Japan's AI Initiative data centers are now being measured in gigawatts, making 25 or 50 megawatts look small. Other Hyperion observations, total HPC AI sector spending in 2024 was 60 billion. The overall HPC market is expected to reach close to 120 billion by 2029, doubling over the next five years. Power requirements are growing in a major rate, and all this leads to some difficult questions.

Will investments continue to grow at these rates? When will ROI become important? Is AI at scale safe, and how can we make it safer? Yeah, let me tell you, all of this is further evidence that those who ignore the HPC market do it to their detriment. it is going to continue to grow as some of us have been waving that flag for years and years. So Hyperion also covered quantum computing, estimating the global quantum computing market at $1.07 billion in 2024 with a projected annual growth rate of 27%,

which would put the market at $2.2 billion in 2027 in two years' time. They also mentioned the growing importance of partnerships as suppliers consolidate and products turn into usable systems and full solutions. While access to quantum computers in the cloud seems like a very natural way to try various modalities, Hyperion sees cloud and on-prem to remain neck and neck, at least for the next few years. And when it comes to applications, no surprise, they expect modeling and simulation to stay at number one by revenue and optimization remaining a major use case. Doug, I hear more and more confident narratives from the vendors,

and it's nice that they feel they are getting close. In some cases, they think they are there already, especially for optimization and simple quantum physics kernels. But I have not seen enough details and evidence, so we'll continue to stay close to this area. My, really our projections at Orion X, continue to look at 233 plus or minus two years for actually useful and program. quantum computers, but breakthroughs can happen.

Last week, we mentioned the new big systems at Oak Ridge and Argonne National Labs. An interesting aspect of these systems is the public-private partnership that was included in the announcements. It's a new model where the U.S. government, DOE, and the National Nuclear Safety Administration, and NSA, which is a semi-autonomous agency within DOE, owns the physical system, but the private companies, HPE, AMD, NVIDIA, and Oracle, share in the usage and benefits. The vendors provide the technology, equipment, and even some capital investment, and they do not have an ownership stake in the physical machines, but have some flexibility in how they provide and use the

overall capacity. Oracle, for example, is also providing access to additional AI computing resources via the Oracle Cloud infrastructure. It's a notable topic that came up very briefly during the press conference at SC25 on Monday. Yeah, it's very reminiscent of utility computing concepts some years ago when everyone used the analogy with electricity provisioning to homes and factories. The partnership model aims to accelerate the deployment of these systems by enabling mutual benefit, allowing both the government and private entities to leverage the supercomputers.

Both the DOE and the industry partners have shared access to the computing power and infrastructure. DOE researchers and scientists would use the supercomputers for open research or according to whatever application isolation policies that they must meet, and private partners gain access to the systems to develop, train, and refine their own frontier AI models and software. In addition to ownership and usage rights, there's also data and intellectual property rights,

and those specifics are likely in individual agreements. But the general principle in the U.S. law right now is that fully AI-generated content cannot be copyrighted. Content where human researchers have made substantial, meaningful contributions may be eligible for copyright. And research conducted by government scientists generally in the U.S. is intended for public benefit and advancing science. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on Insidehpc.com and posted on OrionX.net.

Thank you for listening.