Google strikes $30 billion compute deal with SpaceX to use its Colossus Data Center for AI training

In a dramatic cross-industry tie-up that reshapes the AI infrastructure landscape, Google has agreed to pay Elon Musk’s SpaceX a staggering $920 million per month to lease a massive pool of artificial intelligence computing power. The blockbuster contract, disclosed in a fresh Securities and Exchange Commission (SEC) filing, establishes a direct bridge between the tech giant and the rocket manufacturer to help Google keep pace with the overwhelming commercial demand for its corporate generative AI platforms.

The agreement comes at an incredibly strategic moment for both corporations. For Google, the deal provides immediate computational breathing room as corporate adoption of its agentic AI tier, Gemini Enterprise, grows exponentially. For SpaceX, the contract serves as a massive financial catalyst just one week before the aerospace giant is set to execute a historic initial public offering (IPO) on the Nasdaq, where it is targeting an unprecedented $1.75 trillion valuation.

According to the regulatory paperwork, the cloud services contract will formally begin in October 2026 and run through June 2029, representing a total value exceeding $30 billion if it reaches its full term.

Under the conditions of the lease, Google will gain direct network access to an infrastructure footprint encompassing approximately 110,000 Nvidia GPUs, alongside corresponding central processing units (CPUs), memory systems, and localized networking components.

The SpaceX Anthropic deal

The deal is structurally similar to an even larger infrastructure lease SpaceX finalized with Claude creator Anthropic, under which Anthropic pays $1.25 billion per month for exclusive access to the full output of SpaceX’s massive Colossus 1 data center cluster near Memphis, Tennessee.

By adding Google to its client roster, SpaceX has suddenly created an artificial intelligence infrastructure division that generates an annualized revenue run rate of more than $26 billion out of thin air.

The irony of big tech’s hardware bottleneck

The partnership exposes a fascinating irony: Google is widely considered to be one of the most compute-rich companies on Earth, possessing its own massive global network of proprietary server farms and custom-designed Tensor Processing Unit (TPU) AI chips. Parent company Alphabet has already committed a record-shattering $180 billion toward internal capital expenditures this year and recently executed an $80 billion equity sale to fund additional data center construction.

Yet, even with those aggressive financial commitments, building physical data centers from the ground up can take years, creating a critical bottleneck step for immediate deployment.

Google Cloud representatives downplayed any long-term dependency on outside infrastructure, framing the contract as a tactical stopgap. The company noted that the short-term, timely agreement ensures it has the immediate bridge capacity necessary to meet surging customer demand for Gemini Enterprise, which has scaled far faster than internal modelling anticipated.