@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260316

Episode Date: March 16, 2026

- GTC, G stands for growth - Nvidia’s historical and future market segments - Nvidia Nemotron agentic model, LPU team - Chipper clouds: Cerebras+AWS, d-Matrix+Gimlet Labs - AI’s next big thing? AMI’s "world model” JEPA - AMI’s massive $1B+ *seed* round - European AI companies - InsideHPC signs off [audio mp3="https://orionx.net/wp-content/uploads/2026/03/HPCNB_20260316.mp3"][/audio] The post HPC News Bytes – 20260316 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 HBC Newsbytes. I'm Doug Black. And with me, of course, is Shaheen Khan. As the advanced computing world heads into today's Nvidia GTC extravaganza in San Jose. It's not surprising that there's been a slew of AI-related announcements in the run-up to the conference. Now, we've said before, but it's worth repeating the turnout at GTC, the electricity that the event commands, the streaming crowds pouring into the San Jose Sharks professional sports hockey stadium for a standing room only crowd. It's all a reflection of the dominant position, Nvidia has achieved driving the AI revolution. Well, with a market cap of over $4.3 trillion and a stock price that has appreciated 1,300 percent over the past,

past five years alone, NVIDIA's strategy remains focused on growth. And as you dominate one market segment, growth must come from another new segment, which is why we have seen NVIDA expand in hardware, from GPUs and gaming to HPC and AI, to networking, full servers, clusters, equipment for entire data centers, which they call AI factories, investing in neoclouds, and more recently also CPUs for servers, while it is expected that they would have CPUs for PCs and laptops too. Growth in software has included AI libraries, AI models, applications, and quantum computing software. An expansion at the edge has included automotive, driverless car technologies, robotics, and AI-enabled connected devices,

what we used to call Internet of Things that is having an important resurgence now. Expansion has also been global, and especially in sovereign AI, where various countries, and usually their major telecommunications companies deploy massive AI infrastructure locally. Notably, Nvidia has not expanded into people-intensive markets, like services and support, professional services, system integration, and even a lot of the sales activities, all relying on partners to do those things. Software, CPUs, and Edge are large and fragmented markets and ripe for inclusion of AI towards more autonomous operations,

and specially focused on inference, agentic AI, and federated continuous learning. So, Envidio will go after new markets, push its hardware and software technologies everywhere, and compete and collaborate to do so. Now, this is not the weak when any weakness in the AI market would be noticeable, but there are many, many eyes

looking for any hint of such weakness across the supply chain. Right now, everyone sees AI adoption as an imperative for companies and countries, of whether anyone makes money right now. And so they see the market as big and expanding, with room for many other vendors to grow and thrive. Reading the tea leaves in the market is the strategic priority for vendors and supply chain brokers as much as it is for market watchers. Part of Nvidia's growth strategy includes the new Nemotron 3 super, which they say delivers 5x higher throughput for running agentic AI systems at scale. Agentic AI, of course,

course, now the dominant theme in AI. This is an open 120 billion parameter hybrid mixture of experts model optimized for Nvidia Blackwell chips and is designed to address the costs of long thinking and of context explosion that slows autonomous agent workflows. Another piece of NVIDIA news has been reported by Data Center Dynamics. Invitya is hiring to build out a language processing unit team following its $20 billion GROC licensing deal announced late last year. Nvidia has not gone into detail about its plans with technology from GROC, which makes LPU AI inference chips, but details may emerge at GTC this week. In other AI compute news, Cerebrus continues to align itself with major industry players,

announcing last week that Amazon Web Services is deploying Cerebrus CS3 systems, in AWS data centers. Available via AWS Bedrock, the service will offer open source LLMs and Amazon's Nova models. The companies also are collaborating on a new disaggregated architecture that pairs AWS Traneum with Cerebrous Wafers Scale Engine. And AI inference company D-Matrix and Gimlet Labs, an applied AI research and product company, announced that Gimlet is incorporating D-Matrix Coursair accelerators into the Gimlet Cloud alongside traditional GPUs to deliver what they say is a 10x speed up for agentic AI inference workloads. Yes, and to put it in a bit of a context, Nvidia's expansion strategies have become a model for everyone else. Investors want to hear about

systems and clouds, and so chip vendors want to go up the stack, and preferably all the way to their own neocloud partners. The lines between technology layers were not so sharp to begin with, but they are now certainly a lot more blurry than they used to be. AMD bought ZT systems and cut a deal with Sanmina to move into systems and its own neoclout partners. Grok set up its own cloud before the chip side was acquired by Invidia, and Cerebris has a similar arrangement with G42 and Abu Dhabi-based technology firm and fund, and now with AWS. The D-Matrix announcement is in the same vein.

At the same time, cloud providers look for differentiation. and novel architectures and quantum computers are a good way to do that. Shane, AI investment numbers continue to be eye-popping. Just a few months ago, thinking machines lab, formed by former OpenAI executives, raised a seed round of $2 billion. And now we have Turing award winner Jan Lecon, raising $1.03 billion seed round for his AMI Labs company, which will focus on building so-called world models,

which is AI that learns from reality, not just from language. This is a phenomenon we've talked about in previous episodes. In fact, according to a tech crunch article, AMI Lab CEO Alexandra LeBron says it's the next big thing in AI. He predicts that in six months, every company will call itself a world model to raise funding. So what are world models? Their neural networks designed to understand real world dynamics

with physics and spatial properties built into the world. them, and they ingest more than just text-based content. They also take in input data, including text, image, video, and movement to generate videos that simulate realistic physical environments. You could say they're large language models on steroids with their feet planted in reality. Investors in AMI Labs include several VC firms, technology executives, and corporate investment funds, including Nvidia. One billion is quite a bit of money and naturally very rare as a seed line.

In fact, billion-dollar-plus rounds constitute less than 1% of venture rounds in the technology industry, according to crunch-based news. Yet, they represent over 35% of total capital invested. AMI, as in Ami, as in Monami, means friend in French, but is really short for advanced machine intelligence.

It is headquartered in Paris, France, with offices in New York, Montreal, and Singapore. AMI is not building another generative probabilistic AI model, but a new class of non-generative, objective-driven, predictive AI, that they call joint embedding predictive architecture, JEPA. It is rooted in control theory, dealing with the state of reality as what needs to be transformed by a series of planned actions, into a desired state of reality based on an objective. So it is objective driven, predictive, and rooted in control theory. Those are the three pillars. And if they make it work, they could have an advantage over other models. So score one for Europe in the global AI race.

Now, Europe routinely gets criticized for not investing enough in AI, and maybe that is true across Europe. But France, especially, is becoming quite a center of gravity for AI in Europe. and of course the UK, albeit it's not quite in Europe. Mistral AI has been the most prominent European AI firm. It is also based in Paris and has raised over $3 billion so far, focusing on open-weight large language models. Aleph Alpha, based in Heidelberg in Germany,

works on B-2-B Foundation models and has raised some $500 million. Another company, K-Y-U-T-A-I, pronounced Q-T-A-I, also based in Paris, has raised over $300 million, as a nonprofit lab developing open-source real-time multimodal AI models, and the well-known hugging phase has a major U.S. presence, but its roots and a large portion of its research hub remain in Paris. It has emerged as a global center for sharing and collaborating on AI models. Poolside, another generative AI company based in Paris,

recently raised $500 million of its own, with the valuation of $3 billion, which is similar to AMI's initial valuation. There are others, but let's mention one more, and that's Black Forest Labs in Freiburg, Germany, which was founded by the creators of stable diffusion and builds visual generation models and is also valued at over $3 billion. I'll end with an update on the Inside HPC publication, which provided news and analysis to the HPC community for 20 years. Inside HPC closed down last week, despite growth in our readership.

The publication was ad-supported, but sponsorship dropped dramatically over the past year, and that, ultimately, proved insurmountable. I've seen many supportive comments on social media, which all of us at Inside HPC greatly appreciate. Thank you all. But as you can see, I plan to continue to work with Shaheen on the At-HPC podcast and HPC Newsbytes. Thank you, Doug. Inside HPC was inside all of our hearts too. And I am delighted to continue this podcast with you and all of our listeners. Thank you all. 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.