@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20261012

Episode Date: October 12, 2026

- Gartner's AI predictions - AI Costs, autonomy, accountability - Aurora DAOS storage: 800 servers, 30TB/s - DARPA QBI - Quantum startup funding - Thorium nuclear clock - TSMC–GlobalFoundries packaging deal - Rapidus 2nm update - AI in math "Denial of Human Validation" attack! [audio mp3="https://orionx.net/wp-content/uploads/2026/10/HPCNB_20261012.mp3"][/audio] The post HPC News Bytes – 20261012 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. Tech analyst firm Gartner has already put out its predictions for 2027 and beyond. And they have a recurring theme, the operating consequences of AI. By 2030, Gardner expects... 80% of organizations that offer public-facing AI to experience cost-exhaustion attacks, where abusive queries by online attackers create large computing bills.

By 2028, 60% of global 500 companies will put AI FinOps controls directly into inference to govern these costs in real time. The firm also predicts that by 2029, 80% percent, of new applications will be deliberately disposable used for less than a year. It also expects a quarter of global 500 firms to keep refreshing componentized AI products so rapidly that traditional fast follower strategies become obsolete. Now, these are forecasts, of course, rather than measured trends. Yeah, they go on to say that by 2030, more than 10 billion autonomous agents would flood public services with requests, and physical AI would assist 80% of frontline employees at international companies. They also predict that $10 trillion worth of enterprise-owned energy assets

will turn some large corporations into electricity suppliers. Also interesting, Gartner expects insurance companies to shape AI governance as businesses want coverage against AI related liabilities, and CIOs or chief AI officers will become responsible for maintaining auditable records of consequential decisions made by AI and actions taken as a result. Another forecast says most companies deploying AI will need a dedicated function linking its total cost to business value. Argonne National Lab has expanded the Aurora X-Scale supercomputers Dow's storage system to 800 servers and reports transfer rates as high as 30 terabytes per second across thousands of compute nodes.

This supports simulations, AI workloads, and checkpointing. Storage is a very complex area with many approaches and products. DOS is a distributed object store designed for HPC, also offering file or use. interfaces. It distributes data and metadata and can use user space low overhead IO paths. In storage circles, POSIX compliance is mentioned frequently as an important requirement that can also constrain performance. POSIX defines how applications interact with files, but maintaining its semantics across thousands of clients can be expensive. DioS offers familiar file interfaces with some. some limitations, while its native interfaces can bypass much of that overhead.

It solves a lot of problems in storage. It combines distributed metadata with transactional objects, kernel bypass, and high-performance networks. That's attractive when you have thousands of clients issuing small operations, stream data, or write checkpoints. But separating metadata from data has been around for decades. Sun's QFS and SamFS did that. And today, Versity's Scout F.S goes much further with distributed metadata management and sophisticated indexing. Luster has also evolved beyond a single metadata server. But while set up integration, failure recovery, and long-term support for DOWS are all improving, there are mature products out there that have gone through many years of operational engineering to get there. For example, Google introduced DOWS-based

parallel store but is retiring it in favor of DDN-based managed Luster. Google has similar commercial arrangements with NetApp for storage, VMware for virtualization, and Oracle for databases, so that by itself is not so unusual. One possible factor in Google's decision is a weakness in automated multi-level storage tiering in those that makes it hard to tier down to lower cost storage and requires explicit data movement, which in a public cloud environment can run into bandwidth, operational overhead, and associated costs.

We should really mention that the idea of storage tiering or hierarchical storage management goes back to the 1960s. It became a part of the mainframe stack in the 1970s. Cray's data migration facility, DMF, brought it to supercomputing in the early 90s. It was baked into the OS. It was part of the native directory structure. And it did amazing things, like the OS was able to stop an application, go mount an actual tape cartridge through a robot.

arm, find the data, and resume execution where it left off. SGI then integrated it into IREX and XFS, but the functionality moved over to the storage side, right around when the industry moved to Linux. It all ended up at HPE, which carries it on, though it just announced that DMF has reached end of support and will be phased out in 28, end of an era, really. So, yeah, like you said, storage is complex with a long history. That was could yet, be a long-term winner, but the market may adopt its ideas without adopting DOWS everywhere. Thank you, Shaheen, for that trip down memory lane. Yes.

Now, DARPA, the Pentagon's Advanced Research Unit, has moved four more companies into the final evaluation stage of its quantum benchmarking initiative, bringing the total to six. DARPA isn't selecting winners. It's independently evaluating whether any of these approaches can produce quantum computers whose computational value exceeds their cost by 2033. Meanwhile, neutral atom startup Oratomic, which was founded last year and emerged from Stealth in March, announced $775 million in total funding, including a reported 475 million Series B. The reported valuation is $5.4 billion. Neutral atoms already have substantial contend,

including Q-era, atom computing, Pascal, Plunk, and Google, with ambitious error correction and scaling plans. So the field remains crowded with no single modality breaking away. Major engineering milestones remain, and each modality has its own strengths and weaknesses. None has it all yet. DARPA now has practically every major quantum computing modality in its final evaluation stage,

super computing, trapped ions, neutral atoms, silicon spin, photonics, and topological. One might expect that the industry is approaching a stage where investment would start to consolidate, but fundraising and valuations remain extraordinary. Of course, healthy funding helps innovation, but valuations this high can get in the way of necessary consolidation. Two independent teams, one in Vienna and one in Beijing, have demonstrated. illustrated working thorium 229 nuclear clocks. Conventional atomic clocks rely on transitions involving electrons. These use an unusually accessible transition inside the atomic nucleus with a laser locked to it.

They're not there yet with more precision than the best optical atomic clocks, but the nuclear transition could ultimately support robust, extremely precise timing and new tests of fundamental physics, including searches for dark matter. The breakthrough here is a functioning nuclear frequency reference, and it can be viewed in our recurring theme, great progress, but a lot remains to be done. Quantum technologies typically get split into quantum computing, quantum communications, and quantum sensing or measurement. Quantum computing is the least mature, but with the most promise, quantum sensing has been

an actual market for years. since it is possible to sense changes at the atomic level and use that for precision measurement. That can be used for navigation, synchronization, and scientific discovery. So this is a good reminder that quantum tech extends well beyond quantum computing. TSMC is signed a five-year, $2 billion agreement under which global foundries will make silicon interposers at its Malta, New York factory, with volume production expected to ramp up in the first half of 2028. An interposer is a thin wiring layer of silicon placed beneath several chiplets and above the package substrate. It makes up the dense, short connections between processors, high-bend-with-memory, and other components in an advanced package.

Separately, Japan's Rapidus has announced a network of 17 design partners as it works toward commercial 2-nometer production. production. Japan has committed roughly $15 billion to the effort, but Rapidus still needs customers, competitive yields, and reliable volume manufacturing. So a key question here is whether the TSM agreement would pull global foundries back into the leading edge. It certainly takes it into high-end packaging, but would leading-edge transistor fabrication follow? Well, not necessarily. interposer's need sophisticated processing and high-density connections, but they don't require the same kind of advanced fab technology. Nevertheless, I think this is a significant pull towards the high-end.

Right now, you could say Global Foundries is capturing parts of the high-end computing without actually moving towards the leading edge of transistor fabrication. Its moves into silicon photonics and advanced interconnects play to its strengths, since these technologies don't require leading-edge transistor nodes. Rapidus, on the other hand, is taking that harder route in logic fabrication, aiming to establish a national capability with the associated backing of the Japanese government and several major companies. It's another reminder that advanced chips depend on a network of specialized vendors, spanning logic, memory, interconnect, packaging, and final assembly.

OpenAI has released hundreds of AI-generated mathematical results through a public repository, with formal proofs for many of them. The company says it is consulted a mathematics advisory group at the Institute for Advanced Study and has supplied information about its methods and computation. The response from the math community has been mixed. Researchers see potentially valuable results across several fields, but critics question the quality of vetting, how prior work is credited, and unequal access when frontier AI models remain proprietary.

computer checkable proofs can establish whether an argument follows from formal assumptions. They don't automatically establish novelty, significance, or good attribution. One big constraint right now is the capacity of the mathematics community to absorb the output. As we all know, AI can spew out material faster than anyone can read it, let alone mathematical proofs that require specialized skills and careful consideration. This is reminiscent of the denial of service attacks we see on websites. I call this a denial of human validation attack. The mathematics could be perfectly valid, even novel,

but a large enough volume of results can overwhelm the scarce expert attention needed to assess and understand them. And even if the results are correct and novel, it's not clear whether the problems being solved are important or interesting enough to justify the validation effort. We could see the same problem in other fields of science, of course. 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.