A consortium anchored by BlackRock and featuring Nvidia equity participation closed a $40 billion commitment to acquire and develop AI-optimized data center portfolios across North America and Europe. The transaction, finalized through a newly formed special purpose vehicle, marks the largest single institutional deployment into physical compute infrastructure on record and signals a structural shift in how allocators are pricing sovereign-grade AI capacity.
The deal encompasses fourteen existing hyperscale facilities and committed capital for nine greenfield builds, targeting 2.8 gigawatts of total power capacity by 2027. Apollo Global Management entered as the debt arranger, structuring $22 billion in senior secured notes against long-term lease commitments from undisclosed cloud tenants. Nvidia's participation—reportedly a $1.2 billion equity stake—grants the chipmaker preferential deployment rights for its Grace Hopper and Blackwell architectures, effectively locking in offtake before construction begins. BlackRock's Infrastructure Partners arm contributed $18 billion in equity, its largest single-strategy deployment since the 2021 Global Infrastructure Partners acquisition.
This is not data center speculation. The consortium structured the vehicle to capture margin compression in public cloud services while demand for private AI training environments accelerates. Public cloud providers have historically owned their infrastructure; this deal introduces a landlord layer between compute buyers and physical assets, creating a new rent-extraction point in the AI value chain. For allocators, the significance is cashflow duration: lease terms average twelve years with annual escalators tied to power costs plus 4.5%, producing inflation-hedged yields in the mid-teens on levered equity. The structure mirrors cell tower sale-leasebacks that reshaped telecom economics in the 2010s, except the underlying demand curve is steeper and the replacement cost for cutting-edge facilities continues to rise.
Second-order effects matter more than the headline figure. Nvidia's equity stake aligns hardware roadmaps with physical infrastructure timelines, reducing deployment friction for enterprise clients who need guaranteed capacity at known dates. For sovereign wealth funds and pension systems watching from the sidelines, this transaction establishes pricing for AI infrastructure as an asset class—expect follow-on bids in the $8-15 billion range within six months as Middle Eastern and Asian allocators compete for remaining quality assets. The deal also pressures hyperscalers to either accelerate their own builds or accept that a portion of next-generation capacity will sit on third-party balance sheets, introducing lease expense where capital expenditure previously dominated.
Operators should track three developments: First, whether Microsoft or Google announce sale-leaseback transactions of existing facilities within 90 days, signaling acceptance of the new capital structure. Second, whether regional utility commissions approve the 340 megawatts of incremental load requested for the Texas and Virginia sites by Q3 2025, as grid constraints remain the binding limitation on deployment speed. Third, whether Apollo syndicates portions of the debt stack to insurance portfolios, which would confirm institutional appetite for infrastructure credit at scale and likely tighten spreads on future deals.
The consortium did not disclose tenant names, but power reserved per facility and the Nvidia deployment priority suggest at least two of the hyperscalers are cornerstone lessees. Contract structures of this duration do not get signed without names institutional allocators recognize.