The chip inside your next AI assistant is so complex that traditional design tools can’t even handle it. That’s why Arm turned to Siemens to verify its new AGI CPU, a chip built specifically for the next wave of autonomous, decision-making AI agents.
Arm’s AGI CPU isn’t just another server chip. It’s built on the Neoverse Compute Subsystem (CSS) V3 platform—think of it as a Lego set for hyperscalers, letting them snap together high-performance cores, memory controllers, and high-speed interconnects like PCIe Gen6 and CXL. The problem? A multi-die design this dense is a verification nightmare. You can’t just run a few simulations and call it a day.
The Emulation Engine Room
Siemens’ Veloce Strato CS platform is the heavy lifter here. It’s a hardware-assisted verification (HAV) system—essentially a giant, specialized computer that emulates the entire chip at scale, running real workloads before the silicon even exists. Arm used multiple towers of these machines to verify the AGI CPU from subsystem to full system, catching timing and power bugs that traditional software-based EDA tools would miss.
“The scale and complexity of modern AI compute platforms demand an enhanced level of verification,” said Karima Dridi, Arm’s VP of productivity engineering. Translation: when you’re building chips for gigawatt-scale data centers, a single bug can cost millions.
Software Before Silicon
But verification isn’t just about hardware. Siemens also deployed its Veloce proFPGA CS prototyping platform, which runs FPGA-based prototypes at near-real-time speeds. That means Arm’s software teams can start writing drivers, testing OS kernels, and doing system bring-up months before the first physical chip arrives. In the hyperscaler world, where deployment timelines are measured in weeks, that head start is everything.
The kicker? This isn’t a one-off collaboration. Siemens is making the same verification stack available to Arm’s entire ecosystem—any licensee or SoC designer building on Neoverse CSS can use it. That’s a big deal for the custom silicon arms race in AI infrastructure.
What This Means
The Arm AGI CPU is a bellwether. As AI agents move from chatbots to autonomous systems that make decisions in real time, the chips running them need to be flawless from day one. Siemens and Arm have essentially built a blueprint for how to verify chips that are too complex for conventional methods. For hyperscalers and startups alike, the message is clear: if you’re not emulating at scale before tapeout, you’re already behind.
