@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20260608

Episode Date: June 8, 2026

- Microsoft Build: AI-native software, new chips, quantum progress - Microsoft Maia 200, Cobalt 200, Majorana 2 - Nvidia’s Windows PC push and RTX Spark - Europe’s proposed CADA, AI Act, AI Continent Action Plan - Global AI strategies: US, China, Canada, India, Japan, and Europe [audio mp3="https://orionx.net/wp-content/uploads/2026/06/HPCNB_20260608.mp3"][/audio] The post HPC News Bytes – 20260608 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. Microsoft holds several annual and special events anchored by their Microsoft Ignite event held in the fall that draws some 15,000 attendees and focused on Enterprise Cloud and security. The Xbox game showcase in the summer and surface hardware in the fall, covers consumers, and Microsoft Inspire is for partners. Microsoft Build, held last week and every spring,

is a close second to ignite in strategic importance. It is focused on developers and sets their technical foundation. Microsoft made a slew of announcements at last week's event, and not surprisingly many were centered on agentic AI, and also touched on quantum. On the hardware front, Microsoft used build to reinforce their full-stack story. Maya 200, its second-generation AI inference accelerator, had already been announced earlier in the year,

but Microsoft said it is now running in production and expanding. Maya 200 is purpose-built for the enormous demands of serving the big AI models, such as ChatGPT and Microsoft's own co-pilot at scale. It replaces Maya 100, which the Build MVP-FAST publication says was not widely deployed and delivers 10 petaflops of 4-bit performance. Microsoft claims it offers 30% better performance per dollar than previous hardware and up to 40% better energy efficiency per token. Cobalt 200, the company's arm-based cloud CPU, was said to scale up to 128 cores per virtual machine, provide 50% better performance than the Cobalt 100, has more storage iops and better efficiency

compared with Legacy X86 CPUs, according to Microsoft. Cobalt 200 is being rolled out through new Azure VMs. Microsoft also announced Magirana 2, its next generation, topological quantum chip featuring a new material stack and cubits that the company says are 1,000 times more reliable than their predecessors. Reliability in this case means error correction. With a new chip, Microsoft said it now expects to have a scalable quantum computer by 2029, cutting its original timeline in half. But as you might imagine, Shaheen, there have been doubters and critics of quantum,

including Microsoft's quantum work. Magirana 2 is a 12-cubit configuration and the only actual chip in the physical topological quantum computing modality space, if you can say it like that. Topological quantum computing has always been controversial because it relies on theoretically quote invisible quasi particles known as Magirana zero modes whose physics is complicated and hard to prove. And there have been too many premature breakthroughs, retractions of claims, and fierce debates over experimental data. It has had support from DARPA and Microsoft and Nokia Bell Labs pursue it, which bring massive credibility by virtue of their market position and the

depth of their research, but the area lacks validation right now. Microsoft says it used their new discovery AI tools to help redesign the chips materials stack, replacing aluminum with lead as the superconducting material, and using an updated indium arsenide-based semiconductor stack. That has produced a major improvement in reliability, according to Microsoft, including a, quote, mean-cubit lifetime, unquote, of about 20 seconds. Now, mean-cubid lifetime is not the same as coherence times, but it does show that the lead-based shield works, and it's certainly not the same thing as running useful quantum applications or even standard quantum computing kernels, but it is a meaningful device-level claim if it holds up. Nevertheless, what they did announce was enough for Microsoft to bring forward

their projections of having a scalable quantum computer by four years from 233 to 2029. I think most of the community is skeptical, even as they might cheer them on. On the software side, there were updates to their software infrastructure, development tools, and all of it moving towards AI-native implementation with more built-in automation. On the Windows side, they had an AI and developer-oriented Windows, Windows subsystem for Linux containers, an experimental intelligent terminal, and they showed their Surface RTS Spark dev box built with NVIDIA for local AI development. Microsoft also introduced MAI Thinking One, its own 35 billion parameter reasoning model. model, along with other models for code, image, voice, and transcription.

It shows that Microsoft is building more of its own AI stack and not just relying on Open AI. Security and governance are important than becoming even more so in the world of AI. So that was another major theme. Microsoft talked about sandboxed agents, found re-hosted runtimes, open claw on Windows, scout, intra-based agent identity, and controls for agent behavior. The goal is to make agents governable and auditable as they move into real workplace automation. They also talked about Microsoft Discovery, which we mentioned before, an agentic R&D application

workspace that automate scientific research workflows, initially focused on areas like material science and chemistry. This is kind of like MATLAB. MATLAB uses the Model Context Protocol MCP to let external AI agents write code, So Discovery looks more AI native and skips manual script writing entirely, but in the process also looks like a more closed ecosystem environment. NVIDIA GPUs have been in PCs for decades, and last year, NVIDIA's DGX Spark, Mac minisized desktop-targeted Linux developers, while Jetson Thor did the same for Embedded Edge users. But last week at Computex, NVIDIA announced its most direct.

move yet into the Windows PC processor market with RTX Spark, an arm-based chip that heavily focuses on AI workloads. It brings the same brute force AI processing philosophy that drives data centers to laptops, lots of GPU accelerators and Nvidia's well-established graphics and AI software stack with a new focus on local agentic AI. Headline specs were impressive. RTF Spark combines a 20-core gray CPU with a Blackwell-class RTX GPU, 6144 Kuda cores, 5th-generation tensor cores with Floating Point 4 FP4 support, and up to 100 gigabytes of unified memory all in a Windows PC, with enough performance to run 120 billion parameter models locally, support million token context windows, and,

It'll intensive 3D rendering, edit very large video files, and play high-end games. Not surprisingly, that makes it a, quote, Halo product, unquote, aimed at the very top of the market, rather than ordinary everyday laptops. The Nvidia laptop enters a crowded market. Big players like AMD, Intel, Apple, and Qualcomm all have competitive products. Apple has set a very high bar with its M-Series chips, combining CPU, GPU, NPU and Unified Memory into highly efficient systems. Qualcomm straddles Arm and Windows with Snapdragon X,

bringing its mobile phone routes into PCs, and emphasizing power efficiency, unplugged performance, and Windows on Arm where NVIDIA will play. Apple, Qualcomm, and NVIDIA all use flavors of TSMC's 3-nometer process, while AMD remains a strong X86 competitor using TSMC's, 4-nanometer class products. A newly reinvigorated Intel is the major outlier,

betting that its panther-like chip built on its own Intel 18A process will restore its competitiveness in performance and battery life and that they'll be ready to leap again when its next generation 14A foundry is ready. Ultimately, RTX Spark is a long-term play that aims to expand in VDIA's total available market and pull more of its software ecosystem into the PC area over time. Just last year, NVIDO also announced an investment in Intel and a project to build PC chips together, including future Intel X86 system on chips

with NVIDIA RTX GPU chiplets. So you could say that was another move into the PC processor market, while RTX Spark is the more direct NVIDIA-led arm path. Now, everyday consumers, as you said, do not have, the budget or the need for this level of overkill AI processing power, and especially when Apple and the X-806 incumbents already handle standard tasks efficiently and are not exactly weak in AI either. So RTS Spark is probably not an immediate mass market disruptor, but that's okay. RTS Spark is less a mainstream laptop revolution on day one and more of a beachhead.

This is a foray into a hotly contested market with big players, strong histories, entrenched ecosystems, and deep pockets. Invidio necessarily needs to target a small segment that it has a shot at owning first and then expand from there over time. That segment is the ultra-premium AI performance Windows PC. Developers, creators, AI enthusiasts, workstation-style laptop buyers, and users who specifically, specifically value local model performance and CUDA compatibility. One important technical point is NVLink. RTX Spark uses NVLink internally to connect the CPU to the GPU, but unlike DGX Spark, there was no mention of using it to connect two laptops and turn

them into one. The European Commission has proposed the Cloud and AI Development Act, or KEDA, alongside the broader AI Continent Action Plan. With this proposal, the EU wants to at least triple its data center capacity over the next five to seven years. The related Invest AI initiative seeks to mobilize 20 billion euros for large-scale AI gigafactories and regional AI infrastructure. The proposal is about building more physical capacity and also about control. It introduces data center acceleration zones to speed up permitting an infrastructure buildout while advancing a four-level cloud sovereignty framework for public sector use.

The direction is toward more European ownership, more European-controlled infrastructure, reduced foreign law exposure, and a stronger bias toward open and interoperable technology stacks when they support sovereignty. This marks a shift in Europe's AI strategy. The 2024 EU AI Act focused mainly on software governance, risk classification, and compliance. KADA addresses the physical layer underneath data centers, cloud capacity, frontier compute, sovereign data storage, and the energy systems needed to support them. It also ties AI infrastructure to Europe's green transition.

New data centers are expected to be more energy efficient, better integrated into the grid, and designed with sustainability, cooling, and resource use in mind. Globally, this reflects a distinctly European model, state-directed, rule-based, sovereignty-conscious, and tied to European policy sensibilities. The U.S. pursues rapid expansion with larger, sometimes much larger, investments through private capital, hyperscaler buildout, executive action, and foreign direct investment. Canada uses a mixed public private model, with emphasis on sovereign compute, Canadian governance, and AI safety, but without the same EU-style sovereignty classification, a formal taxonomy that Cato introduces.

China relies on centralized state control, strict data laws, certification of AI models, and heavy industrial policy. Of course, China is also developing indigenous technologies and the ability to operate without Western suppliers if needed. India is focused on digital public infrastructure and domestic technology capacity coming from behind but making rapid progress. Japan is investing in semiconductors, including sub-2-nanameter fabrication, robotics, which they have done for decades, and AI manufacturing. Many other Asian countries are pursuing sovereign AI and cloud strategies, often involving local telecommunications companies, national cloud providers, or government-backed infrastructure programs. As usual, what makes the EU approach distinctive is the attempt to scale AI infrastructure

across multiple countries while tying cloud sovereignty, public procurement, sustainability, and market rules into one common framework. It is a unique European burden to establish alignment among many countries, but eventually that can become an important strength. 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.