@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250818

Episode Date: August 18, 2025

- Will US Government invest in Intel? - How are Chinese AI chips performing for new LLMs? - No slowdown in funding for new AI chip startups - NSF and NVIDIA chip in for Science [audio mp3="https://orionx.net/wp-content/uploads/2025/08/HPCNB_20250818.mp3"][/audio] The post HPC News Bytes – 20250818 appeared first on OrionX.net.

Welcome to HPC Newsbytes, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HBC Newsbytes. I'm Doug Black of Inside HPC, and with me is Shaheen Khan of OrionX.net. We start with the recurring theme of chips and geopolitics, two stories that reinforce, not that reinforcement is needed. high-stakes role of microprocessors in the global rivalries between East and West and between the U.S. and China. Shaheen, you've said that Intel's continuation as a going

concern is a matter of U.S. national security, Intel being a major American chip foundry, and that the company must not be allowed to wither. And it may just be that the Trump administration agrees with you. Reports emerged last week that Intel and the White House are in talks about the federal government taking a financial stake in the company in a deal that supports President Trump's America First manufacturing agenda while also combating dependence on foreign advanced chip manufacturers. This comes after a White House meeting between Intel CEO Liputan and Trump, which itself followed Trump declaring two weeks ago that Tam should resign from Intel because of alleged conflicts of national interest related to China. Shaheen, it's an

amazing turn of events. Intel has gone from being the accused to possibly invested in a matter of days. It also underscores the increasing willingness of the administration to intervene in the technology business, seen in the recent deal between Trump, Nvidia, and AMD, in which the federal government will receive 15% of revenues of sales of Nvidia and AMD chips to China. Well, if you want high-end chip manufacturing in the U.S. by an American company, Intel is your only choice. If you want the most leading edge chip manufacturing and packaging in the US by any company, without having to wait a few years, then Intel is your only chance. Besides what Intel is trying to do in the US, the most leading edge factories start out in Taiwan and then trickle down to the US

and then other places, by which time they are no longer the most leading edge. This is because it is hard to start the next big wave of technology too far from home, but it is also because leading-edge chips give Taiwan the national security advantage, making them critical for the rest of the world and giving everyone incentive to leave them alone. Along these lines, the MIT Technology Review magazine has an interesting article on TSM's role as a deterrent to Chinese aggression against Taiwan. They present two scenarios that could diminish that role with significant implications. One, TSM is expanding its production capabilities overseas. They're building fabs in the U.S. and elsewhere. Technology Review attributes this in large part to the China threat,

but it follows the U.S. Chips Act, and it excludes the very high-end chips, which presumably is what matters most as it relates to China. And two, because China is building out its own chip design and production capabilities, in part due to the prohibition by the U.S. of TSMC exports of advanced chips to China. I think these efforts will likely make progress, but that progress will be slow. would actually be widening. So the article says that if an invasion or blockade of the island were to happen, the global economy dependent as it is on advanced chips manufactured by TSM might just not suffer irreparable damages. Of course, this line of thought was not boosted by a report last week from the Financial Times either. Yes, the Financial Times broke the story on Thursday

that Deepseek encouraged by Chinese government officials to train their upcoming R2 model on Huawei chips encountered chronic problems with the Ascend processor. This after Huawei technical staff worked on-site alongside DeepSeek engineers, they still couldn't attain successful model training runs. If the story is right, and Financial Times usually does a good job of verifying its stories, the upshot is that DeepSeek resorted to going back to using Nvidia chips for the job, Quote, highlighting the limits of Beijing's push to replace U.S. technology, and quote. Regarding the Nvidia H20 and AMD MI308 GPUs that are now allowed for export to China, they look like they deliver roughly the same performance as China's homegrown GPUs.

We've been hearing also that there is pressure in China for companies to use local AI chips and even to cancel orders for American chips that they are now able to get. Chinese officials are reportedly asking for written explanations from companies why they plan to use the NVIDIA and AMD chips rather than local ones. But it's not about just chips. It includes the whole system and software stack for AI training, inference agents, etc. Sticking with GPUs, the information reported that the RIVOS startup is seeking $500 million in funding with the goal of developing AI inference chips to take on NVIDIA. If they're successful, this would bring total funding for the Santa Clara Company, founded in 2021, to 870 million. The number of chip startups and the amount of money raised to maybe get them to be on the short list of alternatives to the main players is just astounding. So it is a sign of the times that so much money is raised for a company that you may not have heard of.

An aspect of a lot of these new chips is that they are based on the Risk Five instruction set. that will increasingly also help with software and system design as well. As usual, the chip would be manufactured by TSM, which together with Samsung is the place to build them, though Intel in the U.S. and Rapidus in Japan are expected to become new options at the high end. There's another story about Nvidia's involvement with the federal government. The National Science Foundation announced a $152 million AI for Science partnership with Nvidia. NSF will contribute 75 million and NVIDIA $77 million to the project, which will be led by the Allen Institute for AI. The intent is to create an open suite of advanced AI models for the U.S. scientific community,

models trained on scientific data and literature. The tools will be designed for researchers and developers to analyze large amounts of data faster, generate code, and visualizations. And they hope connect new insights to past discoveries. But, Jeanne, am I right that this partnership is somewhat unusual that typically agencies like the NSF would award a contract to a company like Nvidia rather than partner with them? Well, yes, in fact, let me just read part of the NSF announcement. Quote, the development of AI technologies is advancing rapidly, but the cost of creating and researching powerful AI models has grown beyond the budgets of university labs and federally

funded researchers. This growing divide limits the topics that academic researchers can explore, despite their historic role in pioneering many of the foundational breakthroughs that power today's AI models, end quote. So it's an interesting move that increases the role of private industry in research, even as it possibly reduces competition, you could say, and echoes some of our previous comments about the cost of, say, a new AI factory in the tens of billions of dollars, dwarfing the $600 million or so that was spent by the Department of Energy on Frontier, which was the first X-Scale supercomputer. For perspective, and another proof that a fraction of a large number is indeed a large number,

the $77 million that Invidio is putting into this project is less than half of a day's revenue based on the company's total of $61 billion in sales last year. Regardless, I applaud the move. Advancing science is critical for global competitiveness and should be supported. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show.

Every episode is featured on Insidehpc.com and posted on OrionX.net. Thank you for listening.