Topic
Infrastructure
Chips, compute, data centers and power: the physical layer of AI.
Briefing · · 5 min read
The week introductory pricing ended: Lambda raises GPU prices, Gemini 4 doubles in December, Fable gets cheaper
Three price resets in ten days, a frontier-scale open-weight release, and more evidence that power, not chips, is what limits new AI capacity. What it means for your 2027 compute budget.
Research · · 3 min read
The same H100 hour costs nearly twice as much depending on who you rent it from
We checked on-demand GPU prices at eight clouds. The spread is wide, the newest chips carry very different premiums, and the H200 is sometimes the better deal.