Google to Rent 110K Nvidia GPUs from SpaceX in $920M/Month xAI Compute Deal

Google has agreed to pay SpaceX roughly $920 million per month for access to about 110,000 Nvidia GPUs housed in data centers SpaceX inherited through its February merger with xAI. The contract runs from October 2026 through June 2029, with capacity ramping up at a reduced fee through September. If SpaceX misses the September 30 GPU delivery deadline, Google can walk away immediately or accept partial capacity at a discount; either party can terminate with 90 days’ notice after the first year. Google says it needs the bridge capacity to handle demand for its Gemini Enterprise subscription, which has outpaced internal forecasts.

The deal lands days before SpaceX’s planned IPO at a targeted valuation north of $1.75 trillion, and it’s the second major leasing arrangement following an Anthropic agreement for the entire Colossus 1 facility in Memphis. The economics underscore why Musk needs the revenue: SpaceX’s AI segment posted a $2.5 billion operating loss on just $818 million of revenue last quarter, against $7.7 billion in AI-related capex. Renting out infrastructure originally built for Grok lets the company monetize stranded capacity while it sorts out a stalled model roadmap and a pending option to acquire Cursor for $60 billion.

The arrangement is notable for its strange-bedfellows quality — Google’s own IPO filing names SpaceX as a competitor in both AI and connectivity, and the two previously had a reversed cloud pact where Google supplied compute to Starlink. It also signals that hyperscaler GPU scarcity is severe enough to push Google past CoreWeave and Nebius and into a deal with a direct rival, reshaping the neocloud competitive landscape.