@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260907

Episode Date: September 7, 2026

- IBM Nighthawk r2 - Diraq at Equinix - From Better QPUs to Better Infrastructure - Agentic AI for Physical Science Experiments - AI and Jobs: Watch Hiring, Not Only Layoffs - Technopolitics: Restrictions and Dependencies [audio mp3="https://orionx.net/wp-content/uploads/2026/09/HPCNB_20260907.mp3"][/audio] The post HPC News Bytes – 20260907 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 as Shaheen Khan. Quantum had several useful developments this past week, each addressing a different part of the problem. IBM launched Nighthawk R2, which, like its predecessor, Heron, remains a superconducting processor, but unlike its predecessor, provides better execution speed and better control through what IBM calls independent high-speed reset capability. All in all, it pushes the system to above 100,000 circuits per second, which is 25 times more than Heron's throughput, while mid-circuit reset

also supports error correction experiments. The UK, meanwhile, opened a 17 million pound program specifically to identify and overcome the bottlenecks involved in integrating quantum computing into AI and HPC workflows. This is another acknowledgement that useful quantum systems will eventually have to plug into an existing computing environment rather than operate as isolated machines. Quantum effect is widely expected to serve as an accelerator handling workloads it is best suited for working alongside and complementing classical HPC systems. In related news, DRAC, another quantum computing company, is installing a silicon spin machine at an Equinix data center.

The system has only eight qubits, so what is interesting is its design, including cryogenics and controls that are to be installed in a data center environment. So what we're seeing is a continued shift in attention from providing qubits towards making quantum systems usable. circuit execution, error correction, software, scheduling, and physical deployment. And HPC remains a very logical early home because heterogeneous computing is already how supercomputers operate and there are actual quantum-ready applications. We also mentioned advances by Kiora and Microsoft last week.

So several roadmaps are looking very interesting in the 2028-29 timeframe versus the previous 2031, 2032 window. I continue to believe a long road and unforeseen barriers remain before we see the kind of quantum computing that can break codes or do something useful and faster than GPUs. The proof is always in the pudding and as they say, the proof of the pudding is in the tasting, but breakthroughs can happen and the industry is projecting a lot of confidence right now. AI for science keeps moving closer to the experiment itself. Anthropic preview had proposed hardware standard

that allows AI agents to operate instruments, such as microscopes, liquid handlers, and robotic arms, while adjusting parameters during experiments and even recovering from some equipment errors. The journal Nature also is published

at review of AI design physics experiments, where the technology moves beyond optimizing a few parameters towards proposing entirely new experimental configurations. And the Allen Institute, University of Washington, and the Fred Hutchinson Cancer Research Center, all located in Seattle,

launched AI biodeign, where models propose biological designs, scientists test them at scale, and the resulting data feeds the next generation of models. That creates a continuous and hopefully productive cycle between computation and physical experimentation. All of that needs an infrastructure layer underneath and associated supercomputers. And a good example is the Lumi AI system in Finland.

Europe has committed some 388 million euros to Lumi AI, which, as the name suggests, has a focus on AI performance, targeting 10 times Lumi's current AI capacity while retaining significant 64-bit capability. So one aspect is numerical formats and the use of low-percision hard. for high precision calculations. That remains a major topic in HPC. We had an exciting and in-depth conversation with John Gustafson, the foremost expert on the topic,

and the author of the book Every Bet Counts. It will be a full-length podcast that will be released shortly. The second topic is speed of discovery. Some of you would remember my virtuous circle diagram, where theory, experiment, and simulation lead to new insights that lead to new data, leads to new or better models, that lead to new experiments and simulation, and on and on. Travel around that circle, the discovery loop, is getting faster.

And probably the most important aspect, the third one, is governance, and nothing makes that visible as well as saying you will use AI to advance biological and medical science. The journal Foreign Affairs points out that many of the capabilities that lower barriers to beneficial biological design can also lower barriers to harmful work. Once AI can interact directly with instruments and physical systems, permissions, provenance, containment, and human authorization become critical, part of the computing architecture itself and not just the policy overlay, very similar to the impact of AI on cybersecurity, which enables new defenses but also new attacks.

Two studies issued by the U.S. Census Bureau give us unusually concrete data on AI and employment. In the business survey, 18% of firms reported using AI in at least one function, rising to 32% when weighted by employment. Among firms actually using AI, employment effects so far are remarkably small. Most reported no AI-related headcount change, and about two-thirds said AI, was being used solely to augment workers' tasks. So despite several years of predictions about wholesale automation, broad company-level evidence still looks more like gradual adoption and changes in how work is performed than widespread elimination of jobs.

Of course, sheen, these studies are retrospective, a snapshot of the current state of things, rather than predictive. The second of the two Census Bureau studies shows where the effect may be appearing first. Among 22 to 24-year-olds in the industry's most exposed to AI, employment fell by about 12% over the 10 quarters after the release of Chad GPT in 2022, primarily because companies hired fewer people rather than firing existing workers. Now, that's an important distinction because labor market disruption doesn't have to arrive as a wave of layoffs. It can arrive quietly through fewer openings and through AI absorbing some of the work normally given to junior employees.

That, in turn, would create a longer-term problem because entry-level jobs are also how people acquire the experience to become senior workers and fewer openings required that other suitable jobs become available. Technology restrictions between the U.S. and China keep moving into additional layers of the stack. Taiwan prosecutors alleged that 74 restrictions, servers using NVIDIA B-300 processors were routed through Indonesia, Japan, and Hong Kong before reaching China, while another 56 were intercepted. At the same time, the U.S. has declared a national emergency over foreign production bulk power

equipment, explicitly linking grid security to the growing electricity requirements of AI, advanced manufacturing, and data centers. And reporting on America's AI construction boom points to continued Chinese supply chain exposure in areas such as electrical equipment, batteries, and optical communications. So even as Washington tries to restrict China's access to advanced compute, the U.S. buildout has dependencies of its own. This is very much technopolitics in action. Restrictions can move the contest versus ending it. Restrict chips and attention shifts towards servers, then networking, power infrastructure, manufacturing equipment, and materials become relevant. In theory, if you keep following that path, you eventually end up with two completely independent

technology spheres. But the global supply chain is complex and a clean separation is probably impossible, certainly in the short to medium term, even though that feels like the path the world is on. There are just too many dependencies and rebuilding every layer would be enormously expensive and time consuming. The big items like advanced semiconductor manufacturing equipment and the high-end chips that they produce have already been the subject of trade restrictions. Adding new layers of technology by either side makes things more complicated. More likely, governments identify a limited number of intolerable choke points, create redundant domestic or allied supply there, and tolerate interdependence elsewhere. The real strategic skill is the

deciding which dependencies are actually worth eliminating. And then there are rare earths, which, as we have said here, are not so rare, but are quite earthy, meaning mining can be messy and separating and processing them at scale is particularly difficult and environmentally challenging. Referencing the journal Foreign Affairs again, it traces how China ended up dominating much of the processing of rare earths and the permanent magnet supply chain and how that has given China a strategic lever of its own. China built that capability and expertise over decades and now has a substantial time and scale advantage that it can leverage. But that advantage is impermanent. The U.S., Europe, Japan, and Australia are investing heavily

to build alternative mining, processing, and magnet capacity. And they can probably reduce this from a critical dependency to a manageable one over the next several years. Replicating China's scale and cost position across the entire supply chain would take considerably longer. 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. Thank you.