Introduction
Welcome back to Laboratory. This week, our spotlight swings to AWS x Anthropic — the scale-up that swaps spec-sheet bragging rights for workload-fit economics and bets big on Trainium. Think multi-gigawatt campuses racing toward go-live, memory-bandwidth-per-dollar as the real metric, and a tighter hardware-software co-design loop that could pull Amazon out of its cloud-AI slump.
In this briefing, we’ll chart:
The Anchor Effect - How Anthropic’s multi-gigawatt commitment can re-accelerate AWS growth and reshape who captures GenAI spend.
Bandwidth Over FLOPs - Why TCO per TB/s beats raw TFLOPs for RL-heavy and token-hungry workloads.
Systems, Not Specs - The shift to Teton PDS/Max and NeuronLinkv3 to close the scale-up networking gap and improve developer experience.