Thoughts on AI, engineering, and building
Practical writing on AI agents, automation, and the engineering behind them, for the people who run the business.
Hardware & Infrastructurehardware-breakdown
Both vendors now sell racks, not chips. A sourced look at Nvidia's Vera Rubin schedule, AMD's Helios and MI500 plans, and what is still vendor claim.
· 5 min read
Hardware & Infrastructurehardware-breakdown
AI systems now ship as liquid-cooled racks of 72 GPUs on one fabric. Published GB300 NVL72 and Vera Rubin NVL72 specs, and what they ask of a data hall.
· 4 min read
Hardware & Infrastructurehardware-breakdown
What Nvidia announced at Computex 2026, which claims are vendor estimates, and what the specs mean for teams planning inference capacity.
· 4 min read
Hardware & Infrastructurehardware-breakdown
NPU ratings from Qualcomm, AMD and Intel compared, plus the memory constraint that decides which models can run locally on a laptop.
· 3 min read
Hardware & Infrastructurenews-analysis
NVIDIA's GTC 2026 announcements sorted into available now, partner availability in late 2026, and roadmap, with performance claims marked as vendor-stated.
· 4 min read
Hardware & Infrastructurenews-analysis
GTC opens next week. Here is what NVIDIA has already said about Vera Rubin and its agent software, which claims are vendor-stated, and what to watch.
· 3 min read
Hardware & Infrastructurehardware-breakdown
AI clusters use two networks: one inside the rack and one between racks. Specs from Nvidia's Rubin platform and the Ultra Ethernet standard.
· 3 min read
Hardware & Infrastructurenews-analysis
Intel, AMD and Qualcomm each announced new AI PC silicon at CES 2026. What the NPU figures mean, what is new at the edge, and how to evaluate it.
· 3 min read
Hardware & Infrastructurehardware-breakdown
Training happens once; serving happens every time someone asks. How Google's Ironwood, AWS Trainium3 and DeepSeek's sparse attention each attack serving cost.
· 3 min read
Hardware & Infrastructurebrief
SK hynix's record quarter and its 2026 supply talks, plus OpenAI's memory letters of intent, show high-bandwidth memory as a constraint on AI capacity.
· 2 min read
Hardware & Infrastructurenews-analysis
Three chip deals in three weeks are priced in gigawatts. Grid forecasts and a record capacity auction show why power, not silicon, now sets the pace.
· 4 min read
Hardware & Infrastructurehardware-breakdown
NVIDIA GPUs, Google TPUs and AWS Trainium compared on published memory and bandwidth figures, and what the differences mean for buyers.
· 3 min read