@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260615

Episode Date: June 15, 2026

- SpaceX IPO. And HPC - Anti-Nvidia Data-Center Startups - Cerebras IPO as a "bet on Nvidia Fatigue”? - PNNL/ASHRAE/NEMA AI Data Center Energy Performance Framework - ISC2026 Conference next week [audio mp3="https://orionx.net/wp-content/uploads/2026/06/HPCNB_20260615.mp3"][/audio] The post HPC News Bytes – 20260615 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 HPC Newsbytes. I'm Doug Black, and with me is Shaheen Khan. Gene, we've hit a new wave of massive technology IPOs, including last week's launch of Elon Musk's, SpaceX, the biggest IPO ever. There isn't a lot we can say that would add to this tsunami of SpaceX commentary that we've heard, I would only say that what impresses me about Musk's undertakings is that as companies take on Cappex intensive challenges that were previously regarded as pretty much impossible, province only of established enormous companies or of the

U.S. government. If you take Tesla, many startups have tried to get into the car business, but no American company is broken into the Ford GM-Preisseler Triumvirate until Tesla came along. As for SpaceX, Its reusable rockets and the Starlink Global Satellite Network are taking on NASA and the big internet companies on equal footing. One Wall Street analyst opined that were on the cusp of moving from an era focused on efficiency to one of plenty, with data centers and solar power plants stationed in space enabling unlimited access to plentiful and clean energy. But, Shane, I'm really intrigued by what you think all of this may mean. Well, there are points and counterpoints about the growth of Tesla and SpaceX and the role of governments and governments

as customers and subsidies and regulations and such. But maybe keeping it focused on aerospace engineering, combustion engineering, AI, low-earth orbit satellite communication, all these scientific engineering heavy disciplines with a lot of HPC components, and then matching the large valuation of the company to those disciplines, maybe that can finally show how important and, shall we say, rewarding HPC should be. So from that standpoint, I certainly welcome the encouragement to the HPC community. Now, we will see how the valuation holds up, since as a multiple of current revenue, it just dwarfs what a lot of current aerospace companies put together are worth. So a lot seems to rest on the assumption that revenue will grow very fast.

very fast and also unaffected by competitive pressure, especially since space, AI, low-earth orbit, all of that are matters of national security, increasingly so, and geopolitics. And other nations' regions could be expected to favor local players, regardless of the usual market advantages foreign vendors might have. The richest companies in the world must expect to take the slings and arrows of criticism, complaint, and competition.

And this truism extends, of course, to the richest company in the world, NVIDIA. In the past month, at least two AI companies have announced anti-NVIDIA intentions as key parts of their market strategies. Last week, Las Vegas-based cloud computing startup, TensorFlow, says it will refuse to use any NVIDIA products, including its GPUs, meaning NVIDIA's GPUs, as part of its cloud offering,

TensorFlowWave, which has raised $1.5 billion, and says NVIDIA controls too much of the AI infrastructure market, declared that it will use hardware and software from AMD, which, by the way, is a major TensorFlow investor. An article in the Wall Street Journal quotes TensorFlow, co-founder and CEO, saying, I don't like buying things from monopolies. You don't have a lot of leverage.

Then we have Cerebris, which went public last month, and is taking aim at NVIDIA in AI model training and inferencing, and there's Majestic Labs AI, which produces chips with vast amounts of memory, and Descartes, whose software makes it easier to switch between computing technologies. Really interesting developments all over the place in the AI world. Big picture, when cloud providers build their own anything, that's at least a long-term strategic threat to merchant vendors of those things. So, accelerator vendors have to protect future demand in the case.

cloud space for their chips because cloud providers are building their own GPUs. And they are doing that by investing in and selling to GPU as a service neocloud vendors. Invidia did that, and it's AMD's turn now that they have full-scale racks. So I continue to see this as partially driven by technology, but substantially driven by protecting future demand for chips. technologically, as accelerators have become bigger and faster and with higher ratios to CPUs, there's been a change to how systems and data centers are designed and operated. You could say accelerator vendors would see that first and could address it directly or had to address it directly.

And when customer demand for accelerators is going through the roof, accelerator companies have all the power to become system vendors and rack vendors, while traditional system and cloud vendors would take a while to build systems and racks and clouds according to the emerging new accelerator-based architecture. At the same time, you could say cloud vendors are watching all of this and thinking, hang on a second, I can build my own accelerator, just like I built my own CPU. And I have the way withal to build the systems and racks and the needed software stack, and I can capture a growing chunk of the market.

And especially if I have to re-architect my systems, anyway. Observing all of that, you could then say that new cloud providers, let's imaginatively call them neoclouds, would note that cloud providers would take a while to complete all that engineering work and therefore see a window to enter the market right away by just using systems and racks directly from accelerator companies. That would fill a gap in the market, serve as a cushion for even traditional cloud vendors who need additional capacity, and provide future demand for racks from chip vendors who would only be too happy to, well, accelerate the whole market. And it all works if demand continues robust and relentless for a bit, which it all did.

You could definitely say that. In keeping with the times in which systems and data centers are measured by their energy usage, Pacific Northwest National Lab, along with the National Electrical Manufacturers Association and Ashray, an international heating, refrigerating, and. Air Conditioning Association, released an AI Data Center Energy Performance Framework, sorry for the bureaucratic language. They say it's a comprehensive set of technical standards and industry best practices to help data center operators in the development of energy infrastructure, including electrical heating and cooling equipment and integrated systems.

We mentioned NEMA in the past, since it sounds like NEMO, a technology brand used by NVIDH. but it is very much something else. This is NEMA with an A, and it stands for National Electrical Manufacturers Association, which was founded in 1926 a century ago, to standardize electrical equipment. There are products out there with the word NEMA in them, like Nematron, as a reference to compliance with its specifications.

So, yes, it's been around for a century and very well known. Ashery stands for the American Society of Heating, refrigerating, and air conditioning engineers, and has been around even longer. Ashtray was founded in 1894 as a professional engineering society. It took its current name after a 1959 merger. It's just huge in the construction business, even foundational, you could say. It sets the global rules for how buildings are heated and cooled and manage energy. It doesn't have legal authority, but its specifications are widely adopted by local governments and form the standards and official building codes, which make them legally binding. PNNL is, of course, the Pacific Northwest National Laboratory,

as you mentioned, in southeastern part of the state of Washington by the Columbia River. It was established in 1965. So just the fact that these three organizations have come together to represent the electronics, the heating and cooling, and the usage of these big systems speaks to the comprehensiveness of the framework. They said the framework was also informed by an industry survey that three organizations conducted that shows a need for collaboration, clear direction, and standardization for managing AI data center power use.

This is a really good point, because AI is super exponential in its growth, and that by itself causes complexity. It also aims to reduce negative impacts on communities surrounding data center developments, which has contributed to the difficulty of finding locations for data centers. The annual International Supercomputing Conference, ISC, is coming up next week from June 22 to 26 at the Congress Center in Hamburg, Germany. It is expected to draw over 3,500 attendees. To make next generation computing research more accessible, the organizers

have significantly lowered tutorial registration fees and offer low-cost, student access, and academic tickets. This is the first year in many that I won't be there in person, but Shaheen and I will, of course, track it remotely and do our usual coverage and analysis of the top 500 list. Yeah, I really looking forward to that. The conference will cover several technology shifts in our post-Mores law era, for example, heterogeneous supergeny. computer design and practical hybrid workflows. This technological evolution is highlighted in the debut of a new community stage and the opening keynote by Dr. Martin Schulz.

ISC is pursuing several projects to reduce the conference's environmental footprint as well, for example, by upcycling all events signage and banners locally and providing integrated public transit incentives. Really nice job there. It's looking like an excellent conference and I hope you all get to experience it. All right, that's it for this episode. 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.