Silicon Feeds
IBM's next-generation Z processor doesn't just run Z; it runs Arm natively too, and at 5.7 GHz that makes it the fastest Arm chip ever made. At Hot Chips 2026, IBM opened day one by unveiling a dual-ISA core that switches between z/Architecture and Armv9.3 on a per-thread basis, alongside 11 cores on Samsung 2nm, 36…
Meta builds its own AI accelerators, and at Hot Chips 2026 it presents the next two: the MTIA 300 for training and the MTIA 400 for generative AI and recommendation. Recommendation compute at Meta grows 3x a year, and GPUs sit idle waiting on memory for those workloads. I also show the first two MTIA generations on…
Agentic AI runs its tools on the CPU, and no vendor tells you how many cores an agent needs. In this keynote at the Design Automation Conference, I compare AMD Venice, Intel Xeon 6, the Arm AGI CPU and Nvidia Vera, then follow the problem down into advanced packaging, CFETs and thermals. You also hear what EDA tools…
It's not every day ASML calls you up for a favor. In this fireside chat recorded live at Hello Tomorrow in Amsterdam, I sit down with ASML CEO Christophe Fouquet to talk Moore's Law, the company's unconventional origin story, and the two decades of costly trial-and-error behind EUV lithography. We also get into the…
South Korea's AI chip scene is booming - DEEPX, HyperAccel, Mobilint, FuriosaAI, and Rebellions are all racing to carve out a niche between the edge and the data center. Ian Cutress and Sally Ward-Foxton give their honest takes on which companies have the traction, funding, and roadmap to actually make it - including…
South Korea has quietly become one of the most active fronts in the AI chip race, and Sally Ward-Foxton and Ian Cutress are on the ground to find out why. In this episode of The AI Hardware Show, they break down five of the country's boldest AI silicon players — DeepX, HyperAccel, Mobilint, Furiosa AI, and Rebellions…
Broadcom's Beast XPU is coming in 2028 with 16 HBM stacks and 8 compute dies on a single package, but how do you actually build something that big? At OCP Taiwan, Broadcom's VP of ASIC products gave a talk on 3.5D heterogeneous "scale-in" packaging, covering face-to-face bonding, 8x higher thermal density, chiplet…
Today's sponsor is https://www.amtechmicro.com Richard Ho leads hardware at OpenAI. He joined Google's TPU program as one of its first engineers, built Anton supercomputers at D. E. Shaw before that, and now runs the team behind Jalapeño. He tells me how close he came to a blank cheque, why the team bet on wiring HBM…
I walk the show floor at TSMC's Open Innovation Platform (OIP) event in the Bay Area. It's a one-day conference where TSMC puts its partners front and centre. We stop by Cadence, Samsung, SK hynix, Synopsys and GUC to see what's next for HBM thermals, co-packaged optics, UCIe 2.0 and AI-assisted chip design. If you've…
Forget the data center - the real AI action might be happening in microwatts. Sally Ward-Foxton and Ian Cutress dig into the tiniest AI chips on the market, from analog compute upstarts to ARM's revived Ethos NPU line, and debate why an "AI ASIC" isn't really an ASIC at all. Plus: a mysterious defense-linked startup…
Forget the data center - the real AI action might be happening in a coin-cell battery near you. Ian Cutress and Sally Ward-Foxton dig into six ultra-low-power chip architectures built to run always-on AI for years on almost nothing, from analog processors that skip the ADC entirely to sparse compute engines that only…
Edge AI doesn't need a data center - it's already running in retail stores, laptops, and microcontrollers all around you. In part two of this edge AI deep-dive, Ian Cutress and Sally Ward-Foxton go off-script to talk through seven chip companies - Axelera AI, Kalray, Kinara, EdgeCortix, MemryX, SiMa.ai, and…
We're at the final day of the AI Infrastructure Summit, and the booths come down around me as I walk. You get Dnotita's vector database ASIC back from the fab, SK Hynix mocking up high bandwidth flash, Marvell showing photonics, and a 30 watt box running 120 billion parameters for 2,100 dollars. Ruby Goodall, who runs…
Day two of the AI Infrastructure Summit, where the running theme is everyone building their own silicon. I take you from Qualcomm's AlphaWave custom silicon and Broadcom's 17,000 pin Tomahawk 6 packages to a six slot rack you carry through airport security. Alan from Giga.io explains who buys compute with no data…
Day one of the AI Infrastructure Summit drew 9,000 people and a show floor twice the size of last year. I interviewed David Patterson on stage for the day one keynote, and he points to memory as the next limit for AI systems. On my floor walk I cover new liquid cooling tech, Quake on RISC-V, more chips, and why…