@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260928

Episode Date: September 28, 2026

- Alibaba’s full-stack play - Chinese system players - Chinese EUV lithography? - SiPearl Rhea1 - Europe Gets Its CPU Back - Data center politics - Data center constraints everywhere [audio mp3="https://orionx.net/wp-content/uploads/2026/09/HPCNB_20260928.mp3"][/audio] The post HPC News Bytes – 20260928 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. Alibaba used its APSara conference last week in Hangzhou, China, to unveil the Genwu V-900 accelerator. The company said its next flagship model will have five to ten days. trillion parameters and set a target of more than 20 gigawatts of cloud capacity by 2032. All that came during President Xi Jinping's September 23 to 25th state visit to the U.S. last week held during the U.N. General Assembly Week.

AI was part of the U.S.-China discussions, and the two sides subsequently agreed to establish an AI dialogue. The UN Security Council separately discussed AI and international security with Chinese and US AI companies represented. Alibaba is one part of a much broader Chinese advanced computing effort. Huawei, Baidu, and its subsidiary, Honlon Jin, Chinese Academy of Sciences referred to as CAS, and the Sugan-Higon-Higon ecosystem. And of course, Lineshine, now number one on the top. top 500 list of the world's most powerful supercomputers. At the chip manufacturing level, there is SMIC and increasingly Shanghai-Wali, China's

answers to TSM, alongside efforts in domestic lithography ranging from immersion DUV, meaning deep ultraviolet technology, to a reported extreme ultraviolet or EUV prototype, which to date only Allen-based ASML has and is under an export ban to China. Yeah, getting access to the leading-edge chip fabrication remains the big structural weakness in technology for China. So its strategy naturally seems to be to focus on areas where it does have access to similar capabilities or even parity, and then steadily build domestic know-how. If your processors are behind, throw more processors at the workload. Improve the interconnect and packaging. Squeeze more utilization.

out of them through software. Use CPUs wherever you can so you don't need GPUs. And co-designed the model and machine together to get more efficiency. We covered Lineshine, the CPU-only arm-based number one system on the top 500. This happened a few months ago, which demonstrated that China can scale a domestic architecture to the very top, although potentially at a considerable cost in power and efficiency. In parallel, they have added more players and more redundancy across the stack. Alibaba, Huawei, Baidu, and Haigone provide different compute paths. SMIC and Wally provide more than one advanced manufacturing path, and their efforts in advanced photography target the hardest remaining choke point.

Of course, a newcomer can leverage the accumulated industry knowledge about how ASML did it, and there was a report in Reuters last December that China appears to have built an operational EUV prototype that can generate EUV light. That's one of the many hard steps. And they did that with help from recruited former ASML engineers. That's impressive. And even if they managed to do it at scale and with precision, it will be one step in many. The supply chain for ASML include hundreds of specialist vendors. You might say, what if they gain access to all the technical detail available one way or another? Even that is. very different from building, tuning, and operating a production-worthy EUV system.

It took ASML over two decades to get its first system going, replicating that could be faster, but not by a lot. Niparl, the European Union chip development company, has delivered the first Raya 1 CPU to bowl for integration into the Jupiter supercomputer's CPU partition. Rayo 1 is built on TSM 6 nanometer with 80-Arm Neuverse V1 cores, HBM2E and DDR5 memory, and PCIE, Gen 5 with C6, that is CCIX, connectivity. The ecosystem is also beginning to broaden. Cyporel CPUs are planned for the Italia inference system and the European Domestic Cloud Project OpenCube, working with HPE's European Lab, has delivered a Raya server based on CyProl's reference design called the SEN server, as in the River Sen.

The next generation RIA2 is the more important next step. It's described as a dual chiplet processor with more than 96 Neoverse V3 cores connected to UCIE, with tapeout expected in 2027 and deployment beginning around 28. It is already planned for the CPU partition of Europe's second ex-scale system after Jupiter called Alice Rekoke. Europe also has Intel's advanced manufacturing operation in Ireland, all of which means that the EU is moving ahead, even if at a relatively slow pace, with its ex-scale and technology sovereignty strategies. Ray-O-1 is Europe's first modern foray into the high-end CPU space.

So Cypro has had to do a lot of foundational work before it could build a processor. It had to build the company, engineers, tools, ecosystems at the same time. Those are already hard areas that you have to push so it makes sense not to push too hard on technology. C-6 is one example. It was baked into Ray-O-1 before several industry projects, C-6, Gen Z, Open. OpenCAPI, CXL itself, effectively converged around CXL, with the idea that future CXL specification would blend them all into a single proposed standard. With the emergence of AI workloads and the proposed GPU-focused Ultra Accelerator Link,

UA Link, CXL has morphed into a CPU-focused connectivity standard for CPU-to-CPU, CPU-Mory, and CPU-GPU connectivity. For Rea 1, CCX does the job and lays the foundation for Rayo 2. Now, Rayo 2 is probably already largely baked as well, but I would expect that things like the Neovverse V3 compute architecture, newer vector capabilities, memory hierarchy, and chiplet design to have been frozen before the exact CXL generation. We do know CXL is part of the Rayo 2 environment. So my guess is that if CXL 4DOTO IP and validation fit the tapeout schedule, they'll take it. Otherwise, they'll probably cut and run with CXL 3.2.

I also expect Rayo to establish the beachhead. Some back of the envelope calculation says they could sell several thousand units before Rayo 2 is introduced to carry much more of the volume. Raya 1 is 61 billion transistors on TSM's 6 nanometer tech, which is about 100 million transistors per square millimeter, which makes it about a 600 millimeter square chip, which leads to about a couple hundred wafers. Again, back at the envelope. So not a massive run, but easily big enough to establish volume production credibility. And longer term, as you indicated, Intel's high-end fab in Ireland creates an interesting possibility for future European CPUs to be designed and built entirely in Europe.

Texas delivered the headline when Governor Greg Abbott followed on the heels of New York State in pausing data center construction. Now Governor Abbott has altered new state-issued data center permits while the state audits their impact on electricity, water, and other infrastructure. More than 470 gigawatts of large loads are seeking Texas grid connections more than five times the state's peak demand. Nonetheless, data center construction in North America is actually quite enormous. Across the eight largest American markets, about 7.5 gigawatts was under construction in the first half of 2026, up roughly 25%. And more than 80% was already pre-leased. Northern Virginia alone has about 4.5 gigawatts operating and another 2.4 gigawatts under construction. The same constraints are appearing internationally.

Europe and the UK face increasing power and grid connection constraints. Tokyo and Osaka have power and land limitations. On the other hand, India is rapidly expanding its data center base, and China is simultaneously planning enormous additions to both compute and electricity supply. Of course, stopping the next round of permits does not stop data centers already permitted, financed or under construction. There is enough capacity in that. pipeline to keep construction moving for quite some time. And permitting can restart before it creates a hole in actual deployment. And 470 gigawatts in the queue certainly does not mean 470 gigawatts will

actually get built. Data centers start small and grow with increasing load. There are real issues and visible public concern, so you add that to a specially heated election season and you're bound to get a broad political reaction and a good dose of political theater. Like you said, in addition to Texas, New York imposed a one-year moratorium on new hyperscale data centers. Separately, California has tightened oversight, and Virginia, a really big site for data centers, is giving communities more say over where these facilities go and who pays for the infrastructure. Lots of noise pollution issues and water pricing and power pricing. At the other end of the spectrum, states like New Hampshire really want this stuff.

So New Hampshire legislators followed their state motto and proposed a right to compute law, which would establish a legal framework to protect the private ownership and lawful use of computational resources from heavy-handed government restrictions. It has not become law, but the contrast is interesting. States are starting to decide whether compute is something to constrain, to regulate, or actively pursue and protect. Underneath the politics, the physical constraint is real. Power is becoming part of the computer architecture, and it's all tight supply. Power generation, transmission, substations, transformers, water, land, permits, and the usual, how quickly can you place an order

for servers and receive them? It's a bit of the same picture of the global land. Europe and the UK have grid and energy cost constraints. Japan has power and land constraints. India has to build generation alongside compute. And China may have an advantage in coordinating generation, transmission, and data center construction, but its demand is correspondingly big. I see the scarce resource to increasingly be megawatts delivered on schedule. And the global energy picture has become more complicated by the Middle East War,

which adds cost and energy uncertainty. uncertainty, particularly for import-dependent regions, and that could be an accelerator for an existing problem. 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.