@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250106

Episode Date: January 6, 2025

- Beyond EUV lithography, LLNL, DOE - ASML, GPUs, China, trade sanctions - AI infrastructure build-out - TSMC 2nm Chips in 2025 [audio mp3="https://orionx.net/wp-content/uploads/2025/01/HPCNB_20250106.mp3"][/audio] The post HPC News Bytes – 20250106 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC News Bites. I'm Doug Black of Inside HPC, and with me is Shaheen Khan of OrionX.net. Shaheen, it always amazes me how over the past five or six years, how few people I know were generally well-informed. By that, I mean they're not investors and they're not in the tech industry, but just generally well-informed people don't know who NVIDIA is. When they ask,

is NVIDIA? I reply that it's a company changing your life and the world. I bring this up because the Wall Street Journal ran a piece last week about ASML and their EUV machines, EUV standing for Extreme Ultraviolet Lithography, the machines used by TSMC, Intel, Samsung, and other chip manufacturers to produce the most advanced CPUs and GPUs designed by NVIDIA, AMD, and Intel that are powering HPC and AI. The journal characterized ASML as the maker of, quote, the most indispensable machine in the world, the one tool responsible for all the tech in your life, adding, quote, it's made by a company you've never heard of. I bring this up also because the Department of Energy's Lawrence Livermore Lab recently announced they're leading a four-year, $12 million research project whose mission presents quite a challenge. They do it because DOE did the original research on EUV

way back before ASML was born, when Intel and Philips were involved. The new project at LLNL will test new lasers, big aperture thulium laser or BAT versus carbon dioxide lasers, which is the current industry standard. And it hopes to improve EUV source efficiency by about 10 times. Livermore said that this research could lead to a next-generation, quote, beyond EUV, end quote, lithography system that would help produce chips that are smaller, more powerful, and faster to manufacture while using less electricity. Speaking of ASML, their EUV machines are embroiled in the techno-geopolitical struggle between the West and China. ASML machines have been banned for export to the PRC, along with, of no less a controversial Chinese company than TikTok,

may be about to invest $7 billion to access BlackBull chips outside of China. This would make ByteDance one of NVIDIA's biggest global customers. The move by ByteDance is something other Chinese tech companies have done as well, i.e. taking their computing outside China. Last June, the Information reported that ByteDance was renting NVIDIA's best chips from Oracle for AI computing. Now ByteDance is in discussions with data center operators in regions including Southeast Asia about gaining access to Blackwell chips when they become available in 2025. Renting advanced GPUs is not a violation of sanctioned laws. Data Center Dynamics reports that the Biden administration is contemplating capping exports of advanced AI chips to certain

countries, including the UAE and Saudi Arabia, to further reduce the ability of Chinese companies to access AI chips. Even as we hear comments that AI ROI may be lagging behind expectations and that the coming year is a proving ground for AI payback, we also see big tech companies pouring incredible sums into AI chips and infrastructure. In a blog post last week, Microsoft Vice Chair and President Brad Smith said that in 2025, Microsoft is on track to invest about $80 billion to build out AI-enabled data centers around the world, more than half of which will be in the U.S. It's a staggering sum that includes data centers, power, cooling, system build-out, etc. And it highlights the confidence that the big cloud providers have in their global growth and in

their ability to navigate risks in case the AI business slows down, especially with provisioning power and cooling. The investment is part of a larger three-part Microsoft strategy that also includes the development of what they called skilling programs to support more broad AI adoption. And for the third aspect, exporting American AI to, quote, our allies and friends, bolstering our domestic economy and ensuring that other countries benefit from AI advancements, end quote. Let's end with news that TSMC is on track to ship two nanometer chips. Rumors abound that TSMC is testing a new fab with some 5,000 wafers in Taiwan,

and an additional factory will turn it into mass production in 2025 as scheduled. Other rumors have it that by the end of 2025, TSMC will be hitting 50,000 wafers per month and 120,000 by the end of 2026. We have to assume these will be replicated in the U.S. after completion of their fabs in Arizona and other locations. All right, that's it for this episode. Thank you for being with us. HPC News Bites is a production of OrionX in association with Inside HPC. Shaheen Khan and Doug Black host the show. Every episode is featured on InsideHBC.com and posted on OrionX.net. Thank you for listening.