For AI agents — machine-readable, MCP-native, citation-friendly.

September 2026 recorded 44 completed transactions worth $51.5 billion across 1,142 MW—a 46.7% month-over-month increase in deal count and 62.5% jump in capital deployed relative to August. Anthropic's $35 billion acquisition on September 1st anchored the period, followed by Equinix's $10 billion purchase of KKR-held assets and Nvidia's $3 billion acquisition from OpenAI. The transaction surge masks underlying capacity concentration: Richland Parish (Louisiana) and a single facility designated "One" each command 5,000 MW, while Las Cruces (New Mexico) holds 4,500 MW and Ashburn (Virginia) anchors Eastern U.S. density at 4,296 MW across 161 facilities. Month-over-month deal velocity rose 46.7%, though year-over-year comparison is immaterial (the 4,300% YoY figure reflects a low September 2025 baseline). Global operational inventory reached 20,838 facilities totaling 1.804 TW.

The sharp capital concentration in three megadeals signals hyperscaler consolidation of supply rather than greenfield expansion. Anthropic's $35 billion outlay—equal to 68% of monthly deal value—indicates willingness to acquire existing capacity rather than wait on 369 GW of announced pipeline projects (540+ tracked facilities). Conversely, AI-tool query volume fell 47.4% month-over-month, suggesting reduced retail/small-operator interest in queue-depth intelligence during a period when capacity announcements yield to M&A closure. For the next two quarters, expect continued acquisition velocity targeting sub-5 GW regional clusters in power-constrained geographies (Ashburn, Las Cruces, Richland Parish) while capital queries return only when pipeline projects hit procurement gates. Grid operator queue depth and interconnection timelines will remain the binding constraint on organic supply.

Q3 2026 marked a decisive capital consolidation around hyperscaler-led build-out, with Meta and CoreWeave anchoring $246 billion in M&A activity across 1,070 deals in the 90-day window. The verdict distribution tilted sharply toward risk avoidance—216 of 327 scored markets rated AVOID, versus only 24 BUILD—signaling that traditional colocation and regional markets have ceded momentum to power-constrained, ISO-adjacent microsites. Operationally, 61.5 GW of capacity sits live across 21,540 facilities, but 33.7 GW under construction and 27.9 GW planned indicate the real growth engine is in frontier regions: Midland-Odessa (ERCOT), Upper Peninsula Michigan (MISO), and Williston, North Dakota (MISO) all earned BUILD verdicts despite sparse commercial density, a reversal driven entirely by power and grid proximity rather than colocation brand equity.

This 90-day window confirms hyperscalers have decoupled from traditional PE-backed colocation. Meta's three $50M commitments and CoreWeave's paired $48M tranches are not deals—they are infrastructure claims on power supply itself. The AVOID sweep across Western Europe (Dublin, London, Amsterdam, Frankfurt, Manchester) reflects exhausted grid capacity and regulatory friction; the BUILD concentration in rural ERCOT, MISO, and SPP reflects raw power arbitrage. AWS (14.1 GW) and Microsoft (11.1 GW) now operate at scale indistinguishable from utilities, while traditional operators (Digital Realty: 1.8 GW, Equinix: 4.1 GW) manage legacy portfolios. The next 2–4 quarters will test whether announced capacity (15.8 GW) and planned capacity (27.9 GW) materialize; if supply-chain and PPA delays slip into H1 2027, the power gap widens and BUILD markets become noncompetitive.

Watch Midland-Odessa and Upper Peninsula Michigan for actual shovel-moving data; both earned BUILD verdicts but lack established operator presence, making them execution tests for hyperscaler-led facility models. Monitor European markets—if Dublin and Frankfurt announce new power offtake agreements, the AVOID verdict may reset. Finally, track announced-stage projects (6,521 facilities, 15.8 GW); any material miss on FID or COD timing signals that the build consensus is aspirational, not committed, and will force capital back toward operational assets and mature MISO/ERCOT spine markets.