Google has agreed to pay SpaceX $920 million per month for access to roughly 110,000 NVIDIA GPUs, according to a Form FWP (free writing prospectus) SpaceX filed with the SEC ahead of its IPO preparation. The deal runs from October 2026 through June 2029, a 32-month contract that could net SpaceX approximately $30 billion in total compute revenue.

The agreement covers a ramp-up period through September 2026 at reduced fees. Google can terminate the contract if SpaceX fails to deliver the committed GPU access by September 30, 2026, though a one-month grace period applies. If SpaceX delivers partial capacity, fees scale proportionally.

Google described the deal as “bridge capacity” for “stronger-than-expected demand” for Gemini Enterprise, its AI agent platform for businesses.

Why Google Is Going Outside Its Own Data Centers

Google is one of the largest data center operators in the world. The fact that it needs to rent compute capacity from SpaceX tells you something important: enterprise demand for AI agents is outpacing even the biggest infrastructure builders’ ability to keep up.

This is not a one-off. Anthropic already has a compute agreement with SpaceX’s Colossus data center network. Combined, the two AI providers send SpaceX approximately $2.17 billion per month in compute revenue, or about $26 billion annualised. That figure alone would rank SpaceX’s compute division as a major enterprise technology business.

The GPU shortage is not theoretical. It is the constraint that is shaping every enterprise AI decision right now. Who controls the chips shapes which models get deployed, which platforms grow fastest, and what enterprises pay for AI at scale.

What This Means for Business

1. Gemini Enterprise is winning deals at scale

Google would not pay $920 million per month for infrastructure unless enterprises were actually using Gemini Enterprise agents at a rate that justifies the spend. For business owners evaluating which AI platforms to build on, this is a signal that Google is investing heavily to support enterprise-grade workloads, not just consumer AI.

2. Infrastructure costs are baked into every enterprise AI budget

When your AI vendor is paying nearly $1 billion a month just for compute to serve their platform, those costs flow downstream. Enterprise AI is not cheap, and pricing pressure on AI APIs is partly constrained by what it actually costs to run inference at scale. Businesses should factor infrastructure realities into their AI vendor negotiations and long-term planning.

3. The concentration risk is real

SpaceX will soon operate compute infrastructure relied upon by both Google and Anthropic simultaneously, two of the most widely deployed enterprise AI platforms. That is a concentration of infrastructure risk that enterprise risk and governance teams will want to flag. If SpaceX misses a delivery milestone, two of the most important AI providers could face capacity constraints.

4. Speed of enterprise AI adoption is surprising even the builders

Google framed this as a response to “stronger-than-expected demand.” That phrase matters. Even the companies building the AI platforms did not predict how fast enterprise adoption would accelerate in 2026. Businesses that are still in “wait and see” mode are not just behind the curve — they are becoming outliers.

The Broader Infrastructure Picture

The Google-SpaceX deal sits alongside a wave of infrastructure investment that defines 2026. Big tech collectively committed over $700 billion in AI capital expenditure earlier this year. Microsoft confirmed its MAIA 200 chip is now running Anthropic’s Claude models at scale. Amazon signed a $25 billion compute deal with Anthropic. Google is buying GPU time from Elon Musk’s SpaceX.

None of this is about experimenting with AI. It is about scaling AI into the core of enterprise operations.

For business leaders, the message is straightforward: the infrastructure buildout is not slowing down. The companies that have already deployed AI agents in production are using capacity that will only get more constrained. Getting into production now is easier than fighting for access later.

---

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.