Google is paying SpaceX $11 billion a year to rent GPUs — and the AI industry has officially entered the rental market
Google will pay SpaceX $920 million per month for access to approximately 110,000 Nvidia GPUs housed in SpaceX’s data centres, according to an SEC filing on Friday. The deal runs from October 2026 through June 2029 — that’s nearly 33 months, or roughly $30 billion total.
Let that number breathe for a moment. Google, the company that designs its own AI chips (TPUs) and operates some of the largest data centres on Earth, is paying a rocket company more than $11 billion a year to rent GPU time. Building your own infrastructure isn’t fast enough anymore.
What the deal includes
Per the filing:
- 110,000 Nvidia GPUs plus CPUs, memory, and other components in SpaceX’s data centres
- $920 million per month starting October 2026 through June 2029
- Capacity ramps up through September at a reduced fee
- 90-day termination clause after this year — either side can walk with notice
- Delivery deadline: If SpaceX can’t deliver the committed GPUs by September 30, 2026, Google can terminate immediately or accept reduced capacity at a lower fee
Google Cloud told CNBC the deal is “short-term” and intended 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.”
“Bride capacity” is doing a lot of heavy lifting in that sentence. Three years is not a bridge. It’s a structural commitment.
The follow-on to Anthropic’s deal
This is the second massive compute deal SpaceX has announced. Last month, Anthropic signed an agreement to use all compute capacity at SpaceX’s Colossus 1 data centre in Memphis — over 300 megawatts and 220,000 GPUs.
So SpaceX is now monetising the data centres it built for xAI’s Grok by renting them to two of xAI’s biggest competitors. Elon Musk’s company is building GPU infrastructure with one hand and selling it to his rivals with the other. It’s either the most pragmatic business strategy in AI or the most chaotic. Possibly both.
The IPO timing is not subtle
SpaceX is preparing for its IPO next week at a valuation exceeding $1.75 trillion. The company’s AI segment recorded an operating loss of $2.5 billion on just $818 million in revenue in Q1. Its capex for the quarter was $10.1 billion — with $7.7 billion committed to AI infrastructure.
These compute deals are how SpaceX shows Wall Street that its data centres generate revenue, not just losses. The Google deal alone adds roughly $11 billion in annual contracted revenue. That’s a very different story than “we’re burning $2.5 billion a quarter on AI.”
Is the IPO being shored up with rental agreements to make the numbers look better? Absolutely. Does that make the deals fake? No — Google is paying real money for real GPU capacity. But the timing, weeks before the IPO, is the kind of thing that makes securities lawyers reach for their red pens.
The circular economy gets another loop
Remember how this works:
- Google invested in SpaceX back in 2015 (when it was worth $12 billion — it’s now targeting $1.75 trillion)
- Google is also investing $40 billion in Anthropic, which committed $200 billion to Google Cloud
- Google is now paying SpaceX $11 billion/year for compute
- Anthropic is also paying SpaceX for compute at the same data centres
- Alphabet is selling $85 billion in stock (including a $10 billion Berkshire Hathaway investment) to fund its AI buildout
- Alphabet revised its 2026 capex forecast upward to $180-190 billion
The money is going in circles, and every circle adds another zero. The Financial Stability Board is already warning about private credit exposure to AI. These new deals don’t make that warning less relevant.
Neoclouds feeling the heat
SpaceX’s compute rental business puts it in direct competition with “neoclouds” like CoreWeave and Nebius — companies that built GPU cloud infrastructure specifically to serve AI workloads. CoreWeave and Nebius stocks dropped on the SpaceX-Google news before recovering slightly.
When a rocket company with $1.75 trillion in IPO proceeds can outspend you on GPU procurement and undercut you on pricing because it’s subsidised by a rocket business, your moat looks a bit thin.
What this means for New Zealand
The global compute squeeze is real enough that Google — one of the largest infrastructure builders on Earth — is paying $11 billion a year for someone else’s GPUs. For NZ, this reinforces that the data centre capacity gap isn’t closing — it’s widening.
NZ’s competitive advantage remains renewable power and relative political stability. The Datagrid AI factory near Invercargill (280 MW) could become more attractive as a Tier 2 overflow site if the hyperscalers continue locking up Tier 1 capacity. But these SpaceX deals show that the big players are willing to spend almost any amount to avoid waiting in line for their own data centres.
🔍 THE BOTTOM LINE
Google paying SpaceX $11 billion a year to rent GPUs isn’t a signal of healthy market dynamics. It’s a signal of desperation. When a company that spent $45 billion on capex last quarter can’t build fast enough and has to rent from a rocket company, the infrastructure bottleneck is the story. SpaceX is turning its money-losing AI division into a revenue engine just in time for its IPO, Anthropic and Google are paying whatever it takes to keep their models running, and the compute circular economy keeps spinning. Someone should probably be taking notes on what happens when the music stops.
❓ Frequently Asked Questions
Q: Why doesn’t Google just build more data centres? A: They are — Alphabet revised its 2026 capex to $180-190 billion. But data centres take 2-3 years to build and bring online. GPU demand is growing faster than construction timelines. This deal is “bridge capacity” that costs more per unit than owning, but is available now.
Q: Is SpaceX now a cloud computing company? A: Partially. SpaceX still makes most of its revenue from launch services and Starlink, but its AI compute rental business just landed $11+ billion in annual contracts. That’s a significant new revenue stream — and conveniently timed for the IPO.
Q: What does this mean for NZ? A: The global compute squeeze makes NZ’s renewable power advantage more valuable. Data centre projects like Datagrid’s Invercargill facility become more attractive as Tier 1 capacity in the US and Europe gets locked up by hyperscalers. But NZ also faces higher costs if global GPU prices keep rising.
Q: Should I be worried about the circular money flows? A: The Financial Stability Board is already warning about private credit exposure to AI. When Google invests in Anthropic, Anthropic commits to Google Cloud, and they both pay SpaceX for compute — and SpaceX is about to IPO at $1.75 trillion — the interconnections are deep. It works until growth stops. Then it’s 2008 with better PR.
