Cirrascale at PEARC26 in Minneapolis Booth #506

The benchmark is the visible part. The operational work behind it usually isn't. Getting to an audited MLPerf result takes more than capable hardware. It takes planning, integration, validation, and disciplined execution across every layer of the stack. That's what being ClusterReady means. Not a slogan—an operating principle. Because the real measure of AI infrastructure isn't reaching a result once. It's building systems that hit it reliably, repeatedly, on demand. That's the kind of engineering we enjoy. #ClusterReady #MLPerf #MLCommons #AIInfrastructure #AMDInstinct #AICompute #MLPerfTraining

New DCIG analysis from Jerome Wendt on the shift most IT teams haven't clocked yet: we're moving from talking about our infrastructure to operating it through conversation — driven by agentic AI and the Model Context Protocol (MCP). The part worth your attention: VergeOS is already doing this. One code base spanning compute, storage, networking, and data protection, exposed through a single MCP interface — which is exactly why DCIG rated it "Very Strong" on cross-domain context. #AgenticAI #MCP #ITInfrastructure #VMwareAlternative #VergeOS

As agentic AI becomes more autonomous, one question is becoming increasingly important: How do we verify that an AI system can be trusted? Discussions around AI security often focus on models, prompts, and guardrails. Those are important, but they all depend on something more fundamental, a trusted execution environment. Technologies like Confidential Computing, AMD SEV, hardware attestation, and trusted execution environments (TEEs) are helping establish that foundation by protecting workloads while they are running, isolating sensitive data, and providing cryptographic evidence that systems haven't been altered. This becomes even more important as organizations deploy AI across on-premises, hybrid, and sovereign cloud environments, where AI agents are expected to access sensitive data, make decisions, and interact with enterprise systems. The next phase of AI security isn't just about protecting the model. It's about proving that every layer the model depends on, from silicon to runtime, is operating in a trusted state. For those interested in confidential computing and the future of agentic AI security, Hugo Romero's keynote is well worth watching. The discussion around 1:24–1:35 provides a practical overview of AMD's approach to protecting data in the Agentic AI era. 🎥 Watch the session here: Confidential Computing Summit 2026 – Protecting Data in the Agentic AI Era https://lnkd.in/g6yNwrgU #AgenticAI #ConfidentialComputing #AMD #AISecurity #HardwareSecurity 

📢 MIPS, by GF will be at multiple industry events this August. As AI tokens migrate to the edge, compute is shifting from centralized processing to real-time decision-making at the point of action. MIPS is leading this shift, bringing software to silicon with RISC-V. Meet with the MIPS team at these events to learn more.  #MIPS #PhysicalAI #AItokens #FMS 

I talk to a lot of infrastructure teams, and cost is increasingly what's on everyone's mind this year. Hardly anyone is building their own model. They're using one from a big provider or running an open model, and the real money goes to running all of it in production, day in and day out. DDN went right at that with Infinia 2.4 at #RAISE in Paris this week. A few things I took from it: ->  Everybody treats #Inference cost as a GPU problem. DDN is treating it as a data problem too. When a GPU sits there waiting on data, you're paying for it to do nothing, and Infinia 2.4 works on exactly that, keeping the KV cache and model files right next to the GPU and feeding them fast. You end up getting more out of the GPUs you already own instead of buying more. -> #SovereignAI is where DDN's on the firmest ground, if you ask me. A good chunk of what's new in 2.4 is the plumbing that sovereign and multi-tenant setups live or die on, real tenant isolation, identity and access controls, and governance for environments where different groups share the same hardware and can't see each other's data. Lots of vendors move data fast. The hard part is doing it with that kind of isolation baked in from the start, and that's where a lot of the pure storage players keep reaching and can't quite deliver. The orchestration crowd has the opposite problem, they don't own the data to begin with. -> What gives DDN the standing to say all this is scale. Their systems already sit under some of the largest #AIInfrastructure deployments in the world, so none of it is running in a demo lab. That matters most on the sovereign side. More and more countries want their own AI they don't have to rent from somebody else's cloud, and a lot of that is still talk. A vendor that can actually stand the infrastructure up, and prove it holds at scale, is in a good spot. For me it comes down to cost and trust. Keeping #AI running in production without the bill getting away from you, and being able to run it somewhere a regulated business or a government is comfortable with. DDN is going after both. Plenty of vendors are making that same pitch right now, but DDN already runs at that scale for real. Alex Bouzari Tod Earhart Sven Oehme Gary Kortye James Coomer Moiz Kohari Amanda R. Lee Moor Insights & Strategy

The latest #Top500 and #Green500 lists highlight the transformative impact of AI on computing infrastructures and software ecosystems. These rankings showcase the growing importance of powerful computing solutions. However, this evolution goes beyond mere performance; energy efficiency and software preparedness are gaining significance. It's inspiring to see @Arm recognized in both charts as they lead the way in advancing AI and supercomputing technology. https://okt.to/6sFmBy

#Axellio's #XpressFS is a file system built specifically for real-time, streaming time series #data. Traditional file systems like #ext4, #XFS, and #ZFS weren't designed for today's high-speed, high-volume data pipelines. XpressFS closes that gap with simultaneous read/write access, built-in buffering, and full utilization of modern NVMe/SSD speed, all at petabyte scale. Key benefits: → Faster I/O than ext4, XFS, or ZFS → Concurrent read/write with built-in buffering → Scalable, predictable performance at massive volume → Lossless, consistent data delivery downstream Built for #AI workloads, RF collection, sensor networks, and any pipeline where data can't wait to be processed. Learn more: https://lnkd.in/gDFvYEZp #DataInfrastructure #NVMe #StreamingData

"Rail-optimized networking gives us the least latency and least hops." See how Core42 uses Cisco infrastructure to power their AI intelligence grid. 🚀 Watch the clip ⬇️ https://cs.co/6049BEoopX

GPUs can be ordered. Power cannot. Every AI company is now chasing the same scarce resource: abundant, reliable energy. HIVE Digital Technologies has spent years securing power across three continents, infrastructure built to support today's workloads and evolve with tomorrow's computing demands. | Frank Holmes

Witness the power of an all-Canadian AI collaboration in action! Bell has unveiled a groundbreaking agreement, merging Bell AI Fabric’s robust data centre and connectivity with Hypertec Group’s cutting-edge AI servers, BUZZ HPC’s advanced GPU infrastructure, and Cohere’s state-of-the-art LLM expertise. This partnership empowers organizations to deploy AI at scale on Canadian-built and operated infrastructure. What potential do you see in this Canadian AI ecosystem? Share your thoughts: https://lnkd.in/eQwkjZsg