Frontier clusters went from 150 MW in 2023 to gigawatt-class in 2025, and the binding constraint is now power rather than chips. Four hyperscalers guide over $300B a year combined — the fastest capex ramp in corporate history.
AWS (Project Rainier) / Anthropic — AWS activated Project Rainier — nearly 500,000 of its own Trainium2 chips across multiple US data centers, one of the world's largest AI compute clusters and the biggest built on custom silicon rather than Nvidia GPUs. Anthropic trains and serves Claude on it (>5x its previous training compute), scaling toward 1M+ Trainium2 chips — a proof point that frontier-scale compute can run on a hyperscaler's in-house accelerators.
OpenAI / Oracle / SoftBank — A $500B, ~10 GW US data-center program was announced, with the first multi-GW campus rising in Abilene, Texas — the largest compute buildout ever committed. In Sep 2026 the programme's constraint showed up at a second campus: Oracle sent a force majeure notice to Blue Owl, developer of Project Jupiter, the $165B Stargate site in New Mexico, after the state land commissioner twice (March and July 2026) rejected right-of-way for the natural-gas pipeline meant to power it, a delay of about six months. Oracle is not leaving as tenant; the notice lets it defer payments if the campus misses its 2028 target. The binding input was gas, not chips. In October Oracle said it was considering trucking compressed gas to run Jupiter's early phases until Energy Transfer's pipeline arrives, reportedly now February 2027; one analyst estimated a large trailer would last about 40 minutes at 100 MW, and delivered gas would cost several times the hub price.
xAI (Colossus) — Colossus scaled to ~200,000 GPUs and gigawatt-class power as GB200 racks came online — the first single site to approach 1 GW of AI compute.
Microsoft / Google / Amazon / Meta — Combined annual capital spending by the four largest US hyperscalers crossed $200B, dominated by AI data centers — the fastest capex ramp in corporate history.
xAI (Colossus) — xAI brought its Colossus cluster in Memphis online with ~100,000 H100 GPUs (~150 MW), built in months — the first single cluster at six-figure GPU scale.
No coverage matches this filter yet.
xAI 
Meta AI 
Microsoft (Azure Quantum) 
Amazon (AWS) 
Oracle