Google agrees to pay SpaceX $920 million a month for AI compute from Memphis-area data centers

Alphabet’s Google will pay SpaceX $920 million per month for AI compute capacity under a cloud service agreement disclosed in SpaceX regulatory paperwork, according to reporting and the filing itself. The deal, described as a bridge for surging customer needs, gives Google access to approximately 110,000 NVIDIA GPUs, along with CPUs, memory, and related components. GPUs, or graphics processing units, are specialized chips used to train and run large AI models, and large suppliers have struggled to expand supply fast enough for the newest systems.

SpaceX’s filing states that Google’s payments begin in October 2026 and run through June 2029, with capacity ramping up through September at a reduced fee. The contract includes a compliance mechanism tied to delivery timing. If SpaceX fails to deliver access to the committed GPU amount by September 30, 2026, then after a one-month grace period, Google may immediately terminate the agreement or accept the number of GPUs provided with a corresponding pro-rata reduction in monthly fees.

The agreement also sets a longer-term exit window. After December 31, 2026, either party can terminate the contract with 90 days’ notice, according to the filing. SpaceX’s disclosure further specifies that the customer retains ownership and intellectual property rights in content, AI models, and related data. In other words, Google is buying access to computing infrastructure rather than claiming ownership of the physical systems or the data used to run its models.

Multiple outlets tied the compute to SpaceX’s Memphis-area operations built for xAI. A Blockspace report described the arrangement as compute capacity at SpaceX’s Memphis data centers, with the infrastructure originally built for xAI’s Colossus campus in Memphis. The Memphis Flyer additionally said the rented capacity is located at SpaceX facilities in Memphis and Southaven. SpaceX did not spell out additional operational details in the short filing excerpt used in the reports.

Google characterized the contract as a short-term bridge rather than a replacement for its own infrastructure buildout. In an email quoted by AFP, a Google Cloud spokesperson said the agreement is intended to ensure “bridge capacity” to meet surging demand for Gemini Enterprise, an agent-focused AI platform for business customers. The framing matches the timing of the deal, which ramps before full October payments begin.

The new arrangement also underscores how hyperscalers are increasingly leaning on third-party compute providers as AI workloads scale. SpaceX recently disclosed a similar arrangement with Anthropic, under which Anthropic agreed to pay xAI $1.25 billion per month through 2029 for compute from the Colossus 1 data center near Memphis. Together, the deals suggest that Memphis-area capacity has become a critical hub for multiple AI ecosystems rather than a single-company enclave.

Alphabet’s own spending plans announcement why it may be willing to pay for compute access on a tight timeline. In an Alphabet investor presentation posted on Google’s blog, management said it expects capital expenditures of $180 billion to $190 billion for 2026 and expects a significant increase in 2027, with most spending directed toward technical infrastructure. In a separate equity-raise press release, Alphabet said the purpose of the proposed $80 billion in offerings includes funding investments in AI compute infrastructure to meet unprecedented customer demand.

Still, key details remain unreported in the excerpted filing and in the early coverage. TechRepublic noted that it is not clear whether the compute translates into expanded GPU inventory for ordinary Google Cloud customers, and it flagged missing information such as hardware location specifics, whether access is dedicated or shared, service-level commitments, per-GPU pricing, and how the rented capacity will be split between internal AI work and customer-facing services. If the compute is used primarily for internal development or tightly scoped applications, the direct impact on broader cloud availability may be limited.