@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260420

Episode Date: April 20, 2026

- DARPA HARQ Heterogeneous quantum architectures - HARQ program, IonQ, memQ - TSMC and ASML bullish on AI demand - Stanford AI Index 2026 insights - Global AI competition [audio mp3="https://orionx.net/wp-content/uploads/2026/04/HPCNB_20260420.mp3"][/audio] The post HPC News Bytes – 20260420 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. News from the quantum sector created a stir last week when startups, IONQ and MemQ, said they have been selected by the Defense Department's Advanced Technology Organization, DARPA, for development of intercom. connects linking multiple types of quantum computers. This is DARPA's HARC system.

This stands for heterogeneous architectures for quantum. The project's goal is to enable a new class of network quantum computers that combine different qubit modalities into configurations that are modular and scalable. Also last week, IONQ said it is linked two separate trapped ion quantum systems via photonic connections. They said this is the first time commercial quantum computers have been directly networked. Both pieces of news boosted IonQ's stock about 50% over the past week. There's a lot of progress in the quantum competing world along those lines with various vendors pursuing similar results. The name of the DARPA program, as you mentioned, is

heterogeneous architectures for quantum arc, and it parallels the evolution and growth of HBC systems with increasing heterogeneity. The use in HPC systems of CPUs and various accelerators, vector, matrix, FPGAs, etc., and different memory and interconnect technologies points to the way to include quantum technologies in those same systems. The program will explore the integration of different qubit technologies, each optimized for computation, memory, or communication and interconnects. So it is heterogeneous in the sense that it aims to use quantum technologies wherever possible. But if successful, it would make it possible to also consider multiple qubit types for just compute. This fits into DARPA's broader strategy, which has shifted from early

research to a focus on utility scale quantum systems, including networking, fault tolerance, and system integration. Globally, similar efforts exist, but with different force and funding and emphasis. The EU and the UK are advancing modular and hybrid approaches through broad ecosystem programs while China is investing heavily in vertically integrated systems and quantum networking. Two chip-related companies that have worked together to achieve mutually beneficial success, we're talking about chip manufacturer TSMC and chip manufacturing equipment maker ASML, both announced Rosie Business Outlook. last week. Starting with Taiwan Semiconductor, the company issued an optimistic report for continued

high demand for AI chips, even in the face of the war in Iran. The company raised its forecast for revenues and downplayed the risk of supply chain disruptions. They also said capital expenditures this year will approach 56 billion that they projected earlier this year, and that of course means more TSM Fabs. And the company reported a gross profit margin of 66% for the first quarter their highest level in more than 20 years. Regarding ASML, the company raised its sales guidance for the year based on strong demand for their advanced chipmaking machines. Among ASML's two largest customers, of course, are TSMC and Intel. Both companies enjoying strong growth due to the AI boom and both companies, along with other chipmakers.

Hustling to secure ASML machines. Yes, ASML has a pretty fantastic backlog as well. The outlook for semiconductor industry has become more bullish. The news from TSM and ASML point to a multi-year AI-driven upcycle, but the outlook is also constrained and includes important caveats. So, the bullish case, one, they see demand accelerating. No talk of a slowdown or a bubble. It's all structural capex lock.

multi-year increase in demand. Two, the supply chain is fully booked. Starting with chips, capacity is pre-sold months and years in advance. It is a seller's market for advanced silicon. Three, it signals growth across the supply chain. ASML is way upstream, where high-end technology supply starts. And end users are at the other end, where demand starts. ASML supplies TSMC, Samsung, Intel, Rapidus,

the very few companies that can use its advanced equipment, which then produce chips for system vendors. So by the time ASML and TSM become bullish, demand signals have gone through the chain and presumably tempered all the way from end users and are to be taken seriously. So what are the caveats? One, physical constraints are the bottleneck. Increasing capacity takes time. Two, geopolitics and trade tensions remain a factor. and especially since the supply chain remains global.

When additional regional capacity is created, it generates resilience, which is a good thing, but it increases costs because it duplicates things. In the context of trade, retaliatory measures can cross industries, so even full independence is not a complete answer. Three, the talk of astronomic valuations and an AI bubble have not quite gone away.

This is especially the case with the profitability of AI software providers. So even if there is full confidence in the long-term performance of the market, there could be short-term setbacks and corrections. But what we can say today is that the high-end sub-5 nanometer chip market is capacity constrained in a structural new wave build-out kind of a way driven by an AI super cycle, but subject to supply limits, geopolitics, and valuation risks. Our listeners have heard us discuss the annual Stanford University AI Index report. The new report is out and it has some disquieting observations about the state of the AI race between the U.S. and the rest of the world.

Shaheen will have more details in his comments, but the report states that the U.S. is only 2.7% ahead of China, as measured by the performance of the top U.S. and top Chinese large language models on a coordinated system. This was the ranking that in 2023 put OpenAI's ChatGPT4-0314 model ahead of China's ChatGLM-6BBI by more than 300 points. A year ago, February, China's DeepSeek R1 briefly tied the U.S. lead model for the first time. And as of last month, Anthropics, Claude Opus 4.6 model scored 1,503 points, while China's Dola seed 2.0 preview scored 1464 points, a difference of 2.7%. Of course, there's plenty more from the AI Index report than just this. Shaheen, what are some of your observations? Well, I look forward to an opportunity to have an interview with the AI Index.

folks, and we've had the pleasure of having Nestor Masley being a guest of this podcast a couple of times in the past three years. The report makes the big observation that AI is now foundational, not emerging, but it is very uneven in how it is understood, used, and govern. AI is a scaling, but the systems to manage it are not keeping up. And that gap is now the central risk and, of course, the opportunity. Let me put it in the list. One, AI capability, is accelerating faster than our ability to measure it in terms of outcome. 2. AI models make bigger improvements on benchmarks within the realm of large language models. AI is now in the economic mainstream.

Costs per token are getting better, so AI accessibility is improving. Enterprise adoption is growing, and investment is at record levels. Next is AI getting adopted in science and research. They mentioned medicine where programs like Alpha Fold have been well-known advances, but they said the progress in AI is also creating validation gaps and expertise gaps that need to be filled and for which we do not appear to be prepared. Without stronger human-in-the-loop systems, you risk having errors, kind of like silent data corruption that are big considerations in IT systems in general.

Next is the geopolitical race, and that's a lot of. in full swing. The U.S. leads, but China and Europe and Asia in general are not far behind, as you mentioned, and they're closing fast. Compute, talent, and data are the new strategic assets, and each one has a different global dynamics. And finally, governance is lagging, as we mentioned. Incidents are rising, transparency is declining, and institutions are struggling to keep pace. The message for the U.S. and Europe from the report is clear and a bit on comfort. It basically says, yes, you're still ahead, but not by as much as you think. Leadership in AI is no longer just about breakthrough models.

It's about who can deploy at scale, and that shifts the game towards infrastructure, not just innovation. At the same time, regulation is becoming a double-edged sword. It is essential, but it has to be just so, or it will push benefits elsewhere. The report makes a case for regulation agility, with the winners being the one, who can regulate dynamically, not just comprehensively. Then there's the real bottom like, compute and energy. The message is AI is no longer a software story, is power grids, data centers, and chips. As Nvidia has been saying all along, AI is now national infrastructure.

And another fault line that is emerging is openness of AI models. It's a strategic choice about control, competitiveness, and global influence. So AI is viewed as an ecosystem competition, and those who align technology, policy, and capacity, the fastest and most dynamically will have the advantage. 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.