SpaceX to Provision 110,000 GPUs to Google in $920M Monthly Compute Lease

Google has signed a 32-month agreement to lease compute capacity from SpaceX’s xAI infrastructure at a rate of $920 million per month. The deal provisions approximately 110,000 Nvidia GPUs alongside associated host processors and memory to support Google’s Gemini Enterprise deployment.

Hardware Allocation and SLA Provisions

Under the terms of the regulatory filing, SpaceX will deliver compute capacity leveraging approximately 110,000 Nvidia graphics processing units (GPUs), integrated central processing units, memory, and associated networking infrastructure. Housed within SpaceX's xAI data center footprint, the physical hardware deployment is scheduled to run from October 2026 through June 2029, following a capacity ramp-up phase through September 2026 at a reduced fee.

The agreement contains strict service-level agreements (SLAs) regarding capacity delivery timelines. If SpaceX fails to provision the committed volume of GPUs by September 30, 2026, Google retains the right to immediately terminate the contract. Alternatively, Google can accept the provisioned hardware at a prorated, reduced fee following a one-month grace period. Beyond the initial setup phase in 2026, the contract permits bilateral termination by either party subject to a 90-day notice period.

Bridging Compute Deficits for Gemini Enterprise

Google is utilizing the SpaceX lease as bridge capacity to satisfy immediate infrastructure demands for Gemini Enterprise, its large-business agent platform launched in October. Despite massive internal capital expenditure—which Google recently revised upward to between $180 billion and $190 billion for the year—the hyperscaler requires external capacity to offset immediate physical hardware bottlenecks. To fund this aggressive infrastructure scaling, Google’s parent company, Alphabet, is executing an $85 billion stock sale, which includes a $10 billion investment from Berkshire Hathaway.

This outsourcing strategy represents a complete reversal of the historical relationship between the two entities. Five years ago, Google provisioned network and compute resources to SpaceX, deploying Starlink ground stations directly within Google data centers to facilitate satellite internet routing. Today, Google must leverage SpaceX's ready-to-use cluster deployments to maintain operational parity with competing large language model providers.

Monetization of xAI Colossus Infrastructure

The deal highlights a major pivot in how SpaceX is managing the data center footprint acquired through its February merger with xAI, a transaction that valued the combined entity at $1.25 trillion. Originally constructed to support training and inference workflows for Elon Musk’s Grok model, the infrastructure is now being aggressively monetized as merchant compute capacity. This shift follows operational setbacks for Grok, including a talent exodus that forced a planned model rebuild and prompted a negotiation for a $60 billion option to acquire the AI coding platform Cursor.

SpaceX’s Q1 financial disclosures reveal the capital pressure driving these lease agreements. The company’s first-quarter capital expenditures reached $10.1 billion, with $7.7 billion dedicated strictly to AI infrastructure. This massive investment yielded just $818 million in quarterly AI revenue, culminating in an operating loss of $2.5 billion for the segment. By leasing its clusters to external entities—having already leased its Memphis-based Colossus 1 capacity to Anthropic in May—SpaceX is establishing predictable, recurring cash flows to support its upcoming public offering, which targets a valuation exceeding $1.75 trillion.