Google has signed one of the largest AI infrastructure agreements disclosed to date, committing approximately $920 million per month to lease AI compute capacity from SpaceX-xAI, according to regulatory filings ahead of SpaceX’s planned public listing.
The agreement will see Google lease roughly 110,000 GPUs, CPUs, memory modules, and related infrastructure components from xAI’s data center platform beginning in October 2026. The contract runs through June 2029 and is estimated to be worth nearly $30 billion over its duration.
Under the terms of the agreement, capacity will ramp up gradually through September 2026 before reaching full deployment. Google retains the right to terminate the agreement if SpaceX fails to deliver the contracted GPU capacity by the end of September, subject to a one-month grace period. Alternatively, Google may accept a reduced number of GPUs with a corresponding reduction in fees.
Google said the deal is intended to meet rapidly growing demand for its AI services.
“The agreement was signed to ensure we have bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected,” a Google spokesperson said.
The contract follows a similar agreement disclosed in May, under which Anthropic committed to leasing approximately $1.25 billion per month of AI compute capacity from xAI data centers over the next three years.
The back-to-back deals signal a dramatic shift in xAI’s business model. Originally established to build and train Elon Musk’s Grok large language model, the company is increasingly positioning itself as a hyperscale AI infrastructure provider, monetizing excess compute capacity through long-term leasing arrangements with major AI developers.
At the center of this strategy is Colossus, xAI’s flagship AI infrastructure campus in Memphis, Tennessee. Alongside the original Colossus deployment, the company is developing Colossus II, which came online earlier this year following land acquisitions completed in 2025.
While Musk has claimed the facility offers up to 1GW of power capacity, independent analysis suggests currently deployed cooling infrastructure may support closer to 350MW, highlighting the extraordinary scale of investment required to support frontier AI models.
The agreements also raise questions about the future economics of AI infrastructure. Rather than building every data center themselves, leading AI companies are increasingly turning to specialized infrastructure providers capable of delivering massive GPU clusters at scale.
For the digital infrastructure industry, the deals underscore a broader reality: AI demand is translating into multi-billion-dollar commitments for power, data centers, networking, and compute capacity. The emergence of xAI as a compute wholesaler suggests a future where AI infrastructure may become a standalone asset class, similar to the evolution of cloud infrastructure over the past two decades.
The scale is striking. Google’s reported commitment alone represents nearly $11 billion annually, while Anthropic’s arrangement could exceed $15 billion per year, making AI compute one of the fastest-growing segments of global digital infrastructure spending.
For Africa, the development serves as another reminder of the widening gap between regions building AI infrastructure and those primarily consuming AI services. While global technology firms are securing hundreds of thousands of GPUs through multi-year contracts, Africa still accounts for a fraction of global data center capacity and AI compute resources. As AI increasingly becomes an infrastructure race, access to power, data centers, cloud platforms, and high-performance compute may prove as important as access to the models themselves.
