@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20251201

Episode Date: December 1, 2025

- CRN's Top 10 Semiconductor Companies of 2025 - Google TPU vs. Nvidia - Neoclouds, New Customer Class - DOE Genesis Mission National Initiative [audio mp3="https://orionx.net/wp-content/uploads/2025/11/HPCNB_20251201.mp3"][/audio] The post HPC News Bytes – 20251201 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 HPC Newsbytes. I'm Doug Black of Inside HPC, and with me is Shaheen Khan of OrionX.net. The CRN publication has taken a look at 10 semiconductor startups developing silicon products, such as chips for AI and networking, these are chips that compete or work with Nvidia chips. The list is a great snapshot of how hot the market is, though I think there are many more chip companies than the 10 mentioned in the article.

A point to be made about all of this is that 10 years ago, the notion of a chip startup was something of a non-starter, even as Nvidia and AMD struggled to gain better positions in the data, center server market against Intel's 90 plus percent market share. But my, Shaheen, how things have changed. Yes, Silicon has definitely come back to Silicon Valley for sure and beyond, with companies springing up all over the world, fueled by AI and the growing importance of feeling sovereign. The list is alphabetical, which is just as well, and it includes Accelera AI, which is a Dutch company working on Edge AI with its second generation chip called

Europa. Number two is Celestial AI. In Silicon Valley, builds optical interconnects and has raised $520 million total from the likes of Fidelity and BlackRock, which indicates how AI investment has gone beyond traditional venture capital firms. Number three is Cornelis Network, a familiar name in HPC. They're based about 40 minutes outside of Philadelphia, and it launched its 800 gigabits per second, CN5,000 scale-out networking family, touting superior message rates, Rocky V2, that's remote direct memory access on converged Ethernet version 2, which speeds things up, and ultra-ethernet. Another Silicon Valley company, D-Matrix, named after a data structure used in machine learning,

has raised $275 million to build what they call digital in-memory computing. Fiorosa AI in South Korea develops energy-efficient AI chips and has raised significant funds also, most recently $125 million in their series C. Their chip is called RNGD and is being evaluated by LD. G, the big South Korean company. Number six is Next Silicon. We covered them a couple of weeks ago. It's based in Israel and launched its Maverick 2 AI chip,

which can adapt to the needs of the applications in real time. Number seven is another South Korean company, Rebellions, which has raised $250 million in its Series C to build chips and systems. Another well-known name with a lot of star power is Tense Torrent, based in Toronto. and with a lot of focus on Risk 5 instruction set. It provides processors, licensed IP, and open source software. It has raised a whopping $1.8 billion so far and is pursuing another $800 million round reportedly.

Another Silicon Valley firm is Savorite's Scalable Intelligence, that's Savorite with a silent T in the beginning, with what it called Omni Processing Unit, which aims to integrate CPU, GPU, memory, and connectivity, in an AI appliance. And finally, X-Sight labs based in Israel, which works on DPUs and Ethernet connectivity. One of the most interesting and potentially impactful tech stories of last week's short week

were reports that meta is in talks to buy AI chips from Google in a move that could result in reducing its meta's dependence on Nvidia. Such a deal could be in the billions of dollars, but nothing has been finalized for. for purchasing Google tensor processing units or TPUs for AI model training and inference. Quote, the biggest story in AI right now is that Google and Nvidia are being extraordinarily competitive. Adam Sullivan, CEO of Data Center Operator Core Scientific, told the Wall Street Journal. They're in a race to secure as much data center capacity as they can. They don't care

about how much revenue they generate. This is about who gets to AGI, artificial general intelligence, first, end quote. Given the size and growth of the AI market and the strong presence of Nvidia in all things AI, the idea that sooner or later there would be a worthy competitor has been out there. Intel missed a boat, and AMD is becoming stronger and stronger.

Cloud providers are doing their own chips that will gradually siphon off workloads, and the smaller upstarts are interesting and very good, but not in the same class yet. In addition to conversations with Meta, OpenAI was reported to possibly access Google's TPU capacity. As for customers, Anthropic actually committed to a multi-billion dollar deal for up to one million TPUs. Google was the first of the cloud providers to do its own AI chip and even run benchmarks, but it always kept it in-house. So if Google suddenly acts like merchant chips and systems vendor, that is a big C-change.

But the move is less dramatic than that. The big news here is the emergence of the so-called neoclouds as an important customer class right above the enterprise. How did that happen? Well, there's intense demand for AI capacity, and neoclouds add a buffer to provide needed capacity and also to help manage risk for all involved, the big cloud providers on one side and end users on the other. They can replicate an entire data center and probably wouldn't mind playing vendors against each other to get a better deal.

Big cloud providers already make chips and systems and data centers. They can sell to new clouds or rent to them, all without having to become a traditional systems vendor. It can all get blurry since it's just another source of capacity in a market that's seemingly insatiable. The Trump administration issued an executive order launching the Genesis mission described as a department of energy led effort to, quote, transform American science and innovation through the power of AI. The effort builds on the integrated research infrastructure, IRA strategy, a national network of scientific supercomputing resources first put forward by DOE in 2020 and advanced by the Biden administration. Now, with the Trump administration carrying the idea forward, an interesting aspect

of the story is that the Genesis mission stands as a rare instance, not only of Trump-Biden-Trump policy continuity, but also a bipartisan support in Washington for the role of leadership class HPC to U.S. national security and economic competitiveness. The imperative that is AI is so widely held that it makes sense for governments to keep reinforcing it. Likewise, pursuit of quantum technologies has emerged as a must. At the same time, science and engineering in HPC have a long and impressive record of fueling strength and prosperity. So it makes great sense to combine and propel these and it's clear that advanced HPC AI quantum technology is recognized across the political spectrum and really across the globe.

It is welcome to use for all who want more investment in science and engineering and putting that investment in the hands of DOE scientists is a great way to matter. maximize its return. 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.