Hyperscalers announced eight-figure AI campus commitments in August 2026, signaling an accelerating race to secure power and permitting before capacity becomes untenable. Meta committed over $9 billion to a 1 GW campus in Alberta and is expanding its Louisiana Hyperion facility to 5 GW; OpenAI announced Project Camellia in Georgia (3.2 GW estimated); ByteDance broke ground on a $38.4 billion, 1 GW-capable facility in Brazil; and Core Scientific doubled leased capacity to 1.1 GW via a 15-year AMD partnership. Combined 2026 hyperscaler AI capex is projected at $775–800 billion, with Google Cloud revenue up 82% YoY and AWS up 37%.
The geography of buildouts is diversifying rapidly beyond the traditional US West regions. Google claimed Wyoming's largest data center (2.7 GW in Cheyenne via Project Tembo); Meta and hyperscalers are betting on Canada, Louisiana, Georgia, and Texas to distribute thermal and grid load. Finland is emerging as a secondary hub, with Pure Data Centres planning an $8.6 billion expansion. However, regulatory friction is increasing: New York paused >50 MW permits pending grid-cost rules, Texas launched a statewide ERCOT queue audit, and Nebraska blocked tax incentives for new data center projects, signaling pushback against unlimited buildout.
For architects: the capex race now splits into two tiers. Tier 1 (hyperscalers with 10+ GW committed) can absorb 18–24 month permitting and construction cycles. Tier 2 (neoclouds, regional providers) must lock power agreements early and execute in 6–18 months to compete. Watch regulatory approval cycles closely—permitting delays and grid constraints are becoming the true bottleneck, not chip supply. Liquid cooling and modular infrastructure are moving from premium to baseline as power densities exceed grid capacity in traditional regions.