Research
AI Infrastructure
The physical side of AI: the chips that do the math, the data centers that house them, the power they need, and the supply chain that builds it all.
6 entries · newest first
Power from space, and the nuclear backup plan
Oct 2026What Musk and Bezos actually propose for powering AI from orbit, how much power space could really deliver compared with one AI campus, and why nuclear is the nearer-term bet.
The AI stack: from app to GPU
Oct 2026Eight layers between the app you type into and the GPU that does the math — agents, model, serving, frameworks, memory, CPU, and GPU, and why memory sets the speed limit.
CPU, GPU, TPU, DPU: the chips behind AI
Oct 2026Four different chips, four different jobs — what each one actually does, and how they pair up inside a real AI server.
How AI Data Centers Get Built
Oct 2026What makes an AI data center physically different, who actually owns these facilities, and why they keep ending up in places you wouldn't expect.
What It Takes to Keep AI Running
Sep 2026A high-level look at the power, water, materials, and hardware churn behind running AI at scale — and what that actually costs.
Silicon to Supercluster
Aug 2026The AI hardware supply chain in six stages — wafer, die, GPU, server, rack, data center — with what each stage is, who builds it, and what it costs.