Google to Pay SpaceX $920 Million Monthly for AI Compute Capacity
Alphabet secures 110,000 Nvidia GPUs in massive $30 billion infrastructure deal
Google's parent company Alphabet has entered into a major agreement with SpaceX to secure massive artificial intelligence compute capacity. The deal, valued at approximately $30 billion over its term, underscores the extreme costs and competition associated with scaling next-generation AI models like Gemini.
Key details
Under the terms of the multi-year contract disclosed in recent regulatory filings, Google will pay SpaceX approximately $920 million per month for access to high-performance computing clusters. The agreement provides Google with 110,000 Nvidia GPUs, alongside the necessary central processing units, memory, and supporting infrastructure.
The arrangement includes a ramp-up period with reduced fees through September 2026. Full-rate monthly payments of $920 million are scheduled to begin in October 2026 and will continue through June 2029. The infrastructure is located within data centers previously operated by xAI, which SpaceX acquired earlier in 2026, including massive clusters such as the Colossus system.
Why this matters
This partnership highlights the staggering financial capital required to maintain a leading position in the AI industry. Even for a company with Google’s immense internal resources, the demand for training and inference capacity has outpaced internal data center expansion, forcing a reliance on external infrastructure providers at a massive scale.
Context
The deal reflects a broader trend of hyperscalers and AI labs securing "utility-scale" compute. SpaceX has rapidly emerged as a dominant infrastructure player following its acquisition of xAI, recently signing a similar large-scale partnership with Anthropic. This consolidation of compute power within a few key providers is reshaping the economics of AI development.
What happens next
Google is expected to begin integrating this new capacity into its Gemini development pipeline immediately. As the ramp-up period concludes in late 2026, the influx of 110,000 additional GPUs will significantly expand Google's ability to train larger models and handle increasing inference loads from its global user base.
Source: Cicero's Published on AI Usage Global, author: AUG Bot
