On May 28, 2026, Anthropic announced a $65 billion Series H funding round at a $965 billion post-money valuation. The round was led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, with participation from Blackstone, Brookfield, General Catalyst and Amazon, among others. Anthropic's annualised revenue had already exceeded $47 billion earlier this month.
The headline numbers are extraordinary. But for anyone working in AI power infrastructure, the detail buried in the announcement is more significant than the valuation.
THE 10 GW NUMBER
Google + Broadcom: 5 GW of next-generation TPU capacity
SpaceX: GPU access in Colossus 1 and Colossus 2
─────────────────────────────────────────────
Total announced: 10+ GW of new AI compute capacity
Existing: Claude available on AWS, Google Cloud, Microsoft Azure
10 gigawatts. In a single funding announcement.
For context: IREN's Sweetwater 1 — the most documented large-scale AI data center project of 2025–2026, with its 12 pre-ordered transformers and 1,400 MW capacity — represents 1.4 GW. Anthropic just announced compute partnerships equivalent to more than seven Sweetwater projects.
WHAT 10 GW OF AI COMPUTE ACTUALLY REQUIRES
Every gigawatt of AI data center capacity requires a specific set of power infrastructure that cannot be improvised:
Grid connection capacity: 1 GW of transmission-level infrastructure
Substations: multiple 225kV/33kV transformation points
MV switchgear: 18–36 month lead time at current demand levels
UPS/battery: backup power for the full compute load
Cooling: liquid or air cooling at unprecedented density
Lead time: 3–7 years from decision to first power (US) · 2–4 years (Europe)
For 10 GW, that means potentially 40–80 large power transformers, gigawatts of transmission capacity, and years of grid connection queue time — at US lead times. The physical constraint does not scale with capital. Anthropic can raise $65 billion. It cannot raise a transformer factory.
THE TIMING — THREE CONVERGING SIGNALS IN ONE MONTH
May 2026 has produced three signals that together define the scale of the power infrastructure challenge:
- Goldman Sachs: $7.6 trillion in AI CapEx forecast 2026–2031, of which $358B is power infrastructure — the critical path for the other $7.2T
- White House Presidential Determination (April 20, 2026): transformers declared essential to national defense; domestic industry cannot meet demand
- Anthropic Series H (May 28, 2026): 10 GW of new compute capacity announced in compute partnership deals
These are not independent signals. They are three descriptions of the same reality: AI compute demand is growing faster than power infrastructure can be built, at any price, in the US.
BLACKSTONE AND BROOKFIELD — THE INFRASTRUCTURE INVESTORS
The participation of Blackstone and Brookfield in Anthropic's Series H is notable beyond the headline. Both are among the world's largest infrastructure investors. Their participation in an AI company's funding round — not in a data center REIT or a utility — signals that the AI/infrastructure convergence is now happening at the capital allocation level.
Blackstone manages over $1 trillion in assets and has been aggressively expanding its digital infrastructure portfolio. Brookfield's infrastructure division has been equally active in data center and power assets. Both know that the value of Anthropic's compute partnerships is inseparable from the power infrastructure that enables them.
Brookfield: global infrastructure fund · data center and power assets
Amazon: 5 GW compute deal · AWS primary cloud partner
Google + Broadcom: 5 GW TPU deal · Alphabet infrastructure
GIC: Singapore sovereign wealth · infrastructure allocation
Temasek: Singapore sovereign wealth · infrastructure allocation
THE EUROPE IMPLICATION
Amazon's 5 GW commitment to Anthropic requires physical data center capacity — racks, power, cooling, connectivity. Amazon's European data center expansion is already constrained by the same transformer lead times and grid connection queues that GridReadiness documents monthly.
Google's 5 GW TPU commitment faces the same constraints wherever those TPUs are physically housed. Both companies have European infrastructure expansion programmes that are directly affected by EU grid capacity and transformer availability.
The 10 GW announced for Anthropic is additional demand on top of the 7 GW of US AI data center capacity already blocked in 2026 (Sightline Climate data). It compounds a supply constraint that the US government has formally declared a national security emergency.
Europe — specifically France — remains the fastest path to powered AI compute for any project targeting commissioning before 2030. The Anthropic announcement makes that path more valuable, not less, because it confirms that demand is real, funded and growing faster than US supply constraints can be resolved.
THE GRIDREADINESS CONTEXT
GridReadiness tracks the power infrastructure layer that makes announcements like Anthropic's Series H physically deployable. The $65 billion is capital. The transformers, grid connections and brownfield sites are the constraint.
For the infrastructure funds and data center developers who participated in or are watching this round — Blackstone, Brookfield, General Catalyst, Insight Partners — the question is not whether AI compute demand is real. Anthropic's $47B annualised revenue and 10 GW of compute partnerships answer that. The question is where the power infrastructure to support that demand gets built, and on what timeline.
"Anthropic has expanded its compute capabilities over recent weeks, signing agreements with Amazon for up to five gigawatts of new capacity, with Google and Broadcom for five gigawatts of next-generation TPU capacity." — Anthropic Series H announcement, May 28, 2026
The power infrastructure bottleneck just got 10 GW bigger. EU transformer slots for 2028 delivery are available now. The window is narrowing with every announcement like this one.