Skip to main content

Google Pays SpaceX $920M Monthly for AI Compute: Why Hyperscalers Are Desperate for GPUs

Anthropic AI data center TPU compute infrastructure
Google is paying SpaceX $920 million per month to rent compute capacity in former xAI data centers — one of the largest infrastructure deals in AI history. The deal exposes a fundamental truth: even the world's largest tech companies can't build data centers fast enough to meet AI demand. Why is Google, which spends $180 billion annually on infrastructure, buying capacity from Elon Musk?

The Deal in Numbers

According to an SEC filing published on June 5, 2026, Google is leasing approximately 110,000 Nvidia GPUs along with CPUs, memory, and other infrastructure from SpaceX. The contract runs from October 2026 through June 2029 at a fixed monthly rate of $920 million. Capacity ramps up through September at a reduced fee.

The terms are strict: if SpaceX fails to deliver the committed GPU count by September 30, 2026, Google can immediately terminate the agreement. After the first year, either party can exit with 90 days' notice.

Why Google Needs External GPUs

A Google Cloud spokesperson confirmed to CNBC that this is "bridge capacity" to meet surging demand for Gemini Enterprise, which Google launched in October 2025. Demand has exceeded expectations.

This is a remarkable admission. Google plans to spend $180–190 billion on capital expenditures this year (revised upward from $175–185B). This week, it announced $85 billion in stock sales — including a $10 billion investment from Berkshire Hathaway — specifically for AI infrastructure. And it's still not enough.

Context: SpaceX's entire AI segment (formerly xAI) reported a Q1 2026 operating loss of $2.5 billion on just $818 million in revenue. Q1 CapEx reached $10.1 billion, with $7.7B committed to AI. Google is paying a company that loses money on AI — but has physical GPUs that Google needs now.

SpaceX: From Rockets to Data Center Mogul

SpaceX merged with xAI in February 2026 in a transaction valuing the combined entity at $1.25 trillion. The following month, Anthropic signed a deal to use all of SpaceX's compute capacity at its Colossus 1 data center in Memphis, Tennessee. Now comes Google.

SpaceX has transformed from a rocket company into a triple-revenue infrastructure platform: launch services, Starlink ISP, and now AI compute leasing. In its IPO prospectus, SpaceX explicitly names Google as a competitor in both connectivity (Starlink vs. Google Fiber) and AI (against Google, OpenAI, Anthropic, Meta, and Microsoft).

Investors are skeptical. Analysts on Hacker News point out that SpaceX trades at approximately 100× revenue (price-to-sales), while the best data center REITs trade at ~10×. "It's a data center REIT bolted onto a social media company bolted onto a launch business," summarized one commenter.

Grok Struggles, Infrastructure Thrives

The irony: SpaceX/xAI's own AI products — the Grok model and chatbot — aren't gaining market traction. Musk admitted in March that Grok needs to be "rebuilt" following a mass talent exodus. The company then secured an option to acquire AI coding startup Cursor for $60 billion.

Meanwhile, SpaceX faces multiple lawsuits and government probes in the U.S. and EU after Grok enabled users to easily create non-consensual deepfake intimate images from photos of adults and children. The EU launched a formal investigation in January 2026.

The paradox: xAI's AI products lose money, but the infrastructure built for Grok now generates revenue from the three largest AI companies in the world. Google, Anthropic, and potentially others are paying for what Musk built for his own models.

European Data Sovereignty Implications

The direct impact on European users is currently minimal — the deal concerns U.S. capacity. But indirect consequences are significant:

  • AI infrastructure consolidation: If even Google — with $180B annual CapEx — must buy capacity externally, smaller European players have no chance of competing without their own data centers.
  • Compute prices decline slower: When hyperscalers pay $920M/month for 110K GPUs, the market price per token stays higher. European startups building on Gemini API will see slower price decreases.
  • EU AI Act and data sovereignty: European companies will increasingly need to consider where their data runs. Capacity in Memphis (Anthropic) or potentially in SpaceX data centers means data outside the EU. Czech data centers (T-Mobile, O2, CEZ) currently lack comparable GPU clusters.
  • Gemini Enterprise in Europe: Google Gemini Enterprise is available to European companies. If Google needs "bridge capacity," it may signal future investments in European data centers — or even greater centralization in the U.S.

Who Wins, Who Loses

Winners:

  • SpaceX/xAI — monetizes infrastructure built for its own AI ambitions. $920M/month = $11B annually from a single deal.
  • Nvidia — 110,000 GPUs in one deal. Each GPU is an H100 or H200 at ~$30-40K. Nvidia revenue from this contract alone: $3.3–4.4 billion.
  • Anthropic — has Colossus 1 in Memphis. With Google as another tenant, infrastructure value increases.

Losers:

  • Neoclouds (CoreWeave, Nebius) — stocks dropped on the announcement. SpaceX is entering the market these players built.
  • Smaller AI startups — when Google and Anthropic hoard available capacity, there's less for everyone else.
  • European data sovereignty — further consolidation of AI compute in the U.S.

Historical Irony

Five years ago, the roles were reversed: Google supplied cloud services to SpaceX for Starlink. SpaceX installed ground stations at Google data centers. Now Google pays SpaceX for compute. The world has turned upside down.

Alphabet originally invested in SpaceX when it was worth $12 billion (2015). Now SpaceX is heading to IPO with a valuation exceeding $1.75 trillion. Google's original investment has grown ~146× — but it's also paying $11 billion annually for compute.

How much will Google pay SpaceX in total for GPU leasing?

The contract is $920 million per month from October 2026 through June 2029 — approximately $31.3 billion over just under 3 years, with an option for early termination after the first year with 90 days' notice.

Why doesn't Google just build its own data centers faster?

Google is spending $180–190 billion on infrastructure this year, but demand for Gemini Enterprise is growing faster than construction allows. Building a data center takes 18–36 months. SpaceX has 110,000 GPUs ready "now" in existing buildings in Memphis. It's a classic build-vs-buy calculation: faster to buy capacity than wait for your own construction.

How does this affect AI service prices for European companies?

Short-term, likely no direct impact. Long-term, compute capacity consolidation among a few players (SpaceX, CoreWeave, hyperscalers) means less competition in the GPU leasing market, which could slow API service price decreases. European companies using Gemini Enterprise should monitor whether Google begins investing in European data centers to meet EU AI Act requirements.

X

Don't miss out!

Subscribe for the latest news and updates.