@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260921

Episode Date: September 21, 2026

- Critical Minerals, AI, and U.S.–China Talks - Open Weights vs. Open Source - Memory: The Next Sovereignty Target - Fujitsu MONAKA for All - DOE’s Science-First Quantum Roadmap for the Nation [audio mp3="https://orionx.net/wp-content/uploads/2026/09/HPCNB_20260921.mp3"][/audio] The post HPC News Bytes – 20260921 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 is Shaheen Khan. In the lead-up to Chinese President Xi's visit to the White House later this week, U.S. Treasury Secretary Bessent and Chinese Vice Premier He Lefeng opened talks in New York yesterday with an economic agenda. that includes extending the trade truce, tariffs, Chinese purchases of U.S. goods, and U.S. concerns about the flow of critical minerals, especially rare earths.

Bestin said the two sides would also discuss AI safeguards, including open and closed-weight models, as well as ways to avoid a deeper split between the two AI ecosystems. So, mines, trade, chips, and AI policy are now sitting in the same. negotiating room, Shaheen. Yeah, there you have it. When technology dominates geopolitics is maybe when you want to call it technopolitics. Rare earths are critical for technology components, but there are other important minerals, so it makes sense to refer to the superset as critical minerals.

We are also seeing growing attention given to both the materials supply chain with minerals that can cause a pause in the manufacturing process and the data supply chain that drives advances in AI and access rights. So it's tariffs, aircrafts, agriculture, and critical minerals in the same bilateral negotiation, but higher level it's about the supply chain for bits and atoms and how they impact commerce and global power dynamics. The open versus closed discussion, meanwhile, is gaining an importance, and more so with the proposed acquisition of Hugging Face by Nvidia, which we discussed a few weeks ago. because governments are now using those terms and negotiations,

even though the industry still does not agree on what exactly, quote, unquote, open should mean for AI systems. So the issue is what exactly is open, as in accessible via a traditional open source download model, and an effort called Open Waldo, spearheaded by Greg Kurtzer of CIQ and Rocky Linux, also a guest on our podcast previously, is trying to turn a stronger version of open source into something operational. Open Waldo's argument is essentially that if we are going to call AI open source, then the source should actually be open. That is a much stronger proposition than downloading a finished model. Open weights are extremely useful. You can run the model yourself, fine tune it, keep your data local and reduce dependence on an AP. provider, but you still may not know exactly how the model was produced.

Under OpenWaldow's stricter definition, open weights is considered only partially open. Fully open source opens all the digital pieces and documents as much of the human process as possible. So think of a trained LLM as the black box people ultimately use. Behind it are four major pieces. First is the code. Second is the training data. training data. Third is the neural network itself, its size, its topology, and the mathematical functions that push information through it. Fourth are the weights, the billions or trillions of numerical values on the network's connections that are adjusted during the training process

until the model produces useful results. There can also be a fifth piece, as you indicated, human intervention during training, people ranking answers, steering behavior, correcting problems, and making development decisions. The effects of such human judgment can be captured in data and weights, but the historical human process itself is harder to save and reproduce. Open Waldo is putting training data, licenses, provenance, tools, training recipes, and resulting weights into a public inspectable chain. Since the inputs and processes are also open,

another group can inspect the provenance, change the training mix, and given enough compute, attempt to reproduce or rebuild the model. I think it's a really important initiative from Greg. Okay, Reuters reports that SK Hynix is in exploratory talks with Intel about manufacturing memory chips in the U.S. for the first time. The options reportedly include leasing part of Intel's delayed Ohio manufacturing complex or forming a joint venture involving Intel and large cloud customers. S.K. Hynix says nothing has been decided and even the type of memory has not yet been specified.

The company is already building a $4 billion HBM facility in Indiana, but that plant is for advanced packaging and testing. The memory wafers will still be fabricated in South Korea and shipped to Indiana. Meanwhile, China's CXMT says its fifth generation DRAM platform has entered mass production, using more aggressive patterning to increase density and lower costs. So memory localization is moving on both sides of the U.S.-China technology divide. Yes, for Intel, an SK-Hinex arrangement could be another boost to its effort to build a merchant semiconductor manufacturing business,

or at a minimum, help put its Ohio manufacturing footprint to productive use, while solving a problem for SK Hinex and really the memory industry as a whole, which is in dire need of additional capacity, which makes the CXMT announcement very interesting. CXMT is already the world's fourth largest DRAM manufacturer and expanding rapidly. Its strength remains conventional and consumer memory rather than the high-end HBM, which could make that additional capacity a pressure release valve for consumer markets, while allowing the top three Samsung, SK-Hinex and Micron, to keep or even increase their focus on high-end, high-margin AI parts. Now, we should mention that memory is generally easier to localize

than CPUs and GPUs because a DRAM chip consists largely of highly regular repeated memory cell structures. HBM adds difficult stacking, packaging, logic, thermal, and yield problems, but conventional DRM does not require quite the same leading-edge manufacturing challenge as the most advanced logic processors. Fujitsu says it will begin selling its 144 core Monaca Arm processor as a standalone product in November alongside its own Monaca servers. This is a change in strategy.

Cloud operators and other server vendors will be able to buy the CPU directly, including companies that compete with Fujitsu's systems business. Fujitsu is positioning the two-nanometer chip for AI inference, enterprise workloads, and HPC. The architecture grows out of Fujitsu's long supercomputing experience, although Monaca is a substantially different arm v9 design. Fujitsu is also explicitly pitching the platform for sovereign infrastructure in Japan and Europe. We have covered Monaca a few times here, including the meaning of its name, so I encourage our listeners to look those up. Monaca is looking like another strong play by the Japanese supercomputing industry. It is also a good example of HPC technologies permeating the broader market, driven by AI applications.

It is definitely interesting that after decades of building processors only for its own systems, Fujitsu is entering the merchant chip business. I think it's a good move. It also adds another serious arm option to a server market that is still dominated by X86 while becoming kind of crowded with several choices of arm, including one from Arm itself, while we all watch Risk Five servers suiting up. Strategically, Japan gets a very high-end domestically designed processor and domestically built server platform for sovereign infrastructure. Monaco will be manufactured by TSM, while everyone is also looking at Rapidist, the Japanese

high-end chip manufacturer that looks like we'll start manufacturing in volume later in 2027 or 28. Even then, there is packaging and the existing complex supply chain, so the whole thing illustrates why full technology sovereignty is difficult. Even a national processor sits inside an international semiconductor packaging, memory, and software ecosystem. We'll end with a quick quantum note. DoE's Office of Science has published a new roadmap aimed at scientifically relevant quantum computing, beginning with competitive grand challenges through 2028. The salient point really is the metric. The OE wants progress judged by verified

scientific results, with hardware and applications co-designed around real problems. So it moves the emphasis from qubit counts and technology advances towards what the machines can actually do. A good move also. 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.