Google has structured $35 billion in AI infrastructure commitments across two distinct transactions: a multi-year cloud partnership with SpaceX and lease-payment guarantees tied to five US data centers Anthropic will occupy for frontier model training. The SpaceX arrangement centers on expanding Google Cloud's compute capacity through SpaceX's satellite communication infrastructure, while the Anthropic financing positions Google as the credit backstop for facilities the AI lab does not own outright.
The SpaceX partnership gives Google access to low-latency satellite connectivity for edge AI workloads and distributed training clusters, reducing dependency on terrestrial fiber routes that competitors control. SpaceX gains guaranteed cloud spend and integration into Google's enterprise AI stack, which already serves 60% of Fortune 500 firms deploying generative models. Neither party disclosed the deal's duration, but procurement documents filed in Delaware reference a seven-year minimum term with renewal options extending through 2039.
The Anthropic component involves Google guaranteeing lease obligations for data centers in Virginia, Texas, Oregon, Iowa, and Georgia—facilities totaling 1.2 gigawatts of power capacity. Anthropic does not own the real estate; it holds long-term operating agreements with infrastructure funds managed by Blackstone and DigitalBridge. Google's backstop reduces default risk for those landlords, enabling Anthropic to secure below-market lease rates without posting security deposits that would drain its $7.3 billion Series C war chest raised in March. The arrangement mirrors Microsoft's 2023 Azure capacity guarantees for OpenAI, which allowed OpenAI to avoid $18 billion in upfront capital expenditure.
The dual deployment signals Google's recognition that model development and inference distribution now require vertical integration across three layers: compute hardware, physical infrastructure, and network transport. Microsoft controls the first two through Azure and its OpenAI partnership; Amazon holds advantages in the third via AWS's global backbone. Google's positioning through SpaceX addresses the transport gap, while the Anthropic backstop locks in a customer for chips Google designs but cannot yet manufacture at the scale Nvidia commands. Anthropic's Claude models run primarily on Google's TPU v5 accelerators, which cost 40% less per inference token than Nvidia H100 configurations but remain scarce outside Google's own data centers.
Operators should monitor three developments by Q3 2026: first, whether SpaceX's Starlink network integrates into Google Cloud regions as a routing option for enterprise customers, which would pressure AWS to accelerate its Project Kuiper satellite timeline; second, whether Anthropic announces a sixth or seventh data center lease, indicating its training runs exceed current capacity and forcing Google to expand its guarantee; third, whether Google converts any portion of its Anthropic lease backstop into equity, a structure OpenAI used with Microsoft in 2023 that triggered $13 billion in mark-to-market gains by year-end.
The combined $35 billion commitment represents 9% of Alphabet's trailing twelve-month free cash flow, deployed without acquiring equity in either counterparty. Microsoft's OpenAI partnership required $13 billion in cash and compute credits for a 49% economic interest. Google is buying option value—on SpaceX's orbital distribution layer, on Anthropic's model differentiation, on keeping both entities out of exclusive arrangements with AWS or Oracle. The next test arrives when Anthropic's current compute allocation runs out, estimated at eighteen months based on disclosed training budgets. If Google declines to expand the backstop, Anthropic will need to raise $4-6 billion in Series D funding or accept dilutive terms from Microsoft or Amazon, both of whom have made overtures since February.