Google will pay SpaceX $920 million a month for AI compute capacity through 2029, using 110,000 Nvidia GPUs and other hardware in SpaceX’s data centers. The deal begins in October 2025, ramps up to full capacity by September 2026, and lets either side terminate with ninety days’ notice after the first year.

SpaceX’s filing warns that if Google fails to secure the committed GPU count by September 30 2026, the company can end the agreement or accept a reduced fee after a one‑month grace period. The arrangement follows SpaceX’s $1.25 trillion merger with xAI and its latest IPO push, positioning the rocket‑maker as a cloud infrastructure competitor to meet escalating demands for large‑scale AI workloads.

Google cited the surge in demand for its Gemini Enterprise agent platform, launched in October, as the driver for the partnership. By leasing compute, Google offsets rising cloud‑spending, which the company now projects between $180 billion and $190 billion for the year. The move also signals Alphabet’s intent to secure steady revenue streams within the tech sector.

SpaceX’s data‑center strategy dovetails with its Starlink satellite network, giving the company a competitive edge over traditional hyperscalers. The partnership follows Anthropic’s use of SpaceX’s Memphis Colossus 1 facilities and precedes SpaceX’s rumored acquisition of AI coding startup Cursor for $60 billion. For Google, the deal provides a bridge to scale services while mitigating latency risks from distributed AI workloads in 2025.