Topic
hyperscaler-capex
7 articles
Signal Briefing: June 20, 2026
FERC moves to fast-track AI data center grid connections, mandating bring-your-own-power or demand flexibility within 90 days — a structural shift in how US grid access is allocated.
Signal Briefing: June 19, 2026
Meta commits 1.6GW to Crusoe in one of the largest hyperscaler-to-specialist AI capacity agreements on record, as FERC simultaneously moves to reshape grid access rules for all U.S. data centers.
The Memory Wall Meets the Concrete Wall: AI's 2026 Capacity Gap
Half of US datacenter capacity planned for 2026 isn't being built, and a parallel DRAM squeeze is exposing the gap between announced AI compute and what physics will actually deliver.
Signal Briefing: June 10, 2026
Broadcom, Apollo, and Blackstone launch a $35B, 20GW XPU platform to serve Anthropic's inference buildout — the largest single private-capital commitment to AI compute infrastructure yet disclosed.
Signal Briefing: June 5, 2026
TSMC's CEO confirms the foundry cannot meet AI demand 'for a long time,' providing the clearest signal yet that chip supply — not capital commitment — is the binding constraint on the AI infrastructure buildout.
The Memory Wall Is the Inference Wall
DRAM at a 15-year high, HBM5 mockups in Taipei, and Altman calling token costs 'a huge issue' all describe the same constraint: memory, not logic, now sets the price of intelligence.
Signal Briefing: May 30, 2026
AWS disclosed that its new random-graph data center network cuts hardware by 69% and power by 40%, and is already the default architecture for most AWS workloads — the most structurally significant infrastructure efficiency disclosure in today's feed.