SpaceX just told Wall Street it has a $40 billion revenue stream locked through 2029, then Musk told Twitter it's more like a month-to-month lease with an escape hatch.
The Summary
- SpaceX's IPO filing revealed a $1.25 billion/month compute deal with Anthropic through May 2029, potentially worth $40+ billion total
- Musk clarified on X that SpaceX only committed to 180 days, with 90-day mutual cancellation after that — and the short-term structure was his request
- The deal positions SpaceX as an AI infrastructure play, not just aerospace, but Musk is keeping his powder dry if xAI or Tesla need the compute
The Signal
The disconnect between what's in the S-1 and what Musk posted matters because it reveals the real power dynamic in AI compute deals right now. Anthropic agreed to pay through 2029. SpaceX agreed to deliver for 180 days. That's not a partnership. That's SpaceX renting out spare capacity until it needs it back.
The Colossus data centers weren't built for Anthropic. They were built because Musk is fighting a three-front war for compute: xAI needs clusters to train Grok, Tesla needs them for full self-driving, and SpaceX itself is layering AI into Starlink routing and Starship autonomy. Anthropic is paying top dollar to use what would otherwise sit idle during ramp-up. The moment any Musk company hits a compute bottleneck, that $1.25 billion monthly check stops mattering.
"The short term was our request, not Anthropic's."
This is the IPO filing as performance art. SpaceX gets to book a massive revenue line that makes it look like a diversified AI infrastructure company, which juices the valuation for public market investors. But Musk is simultaneously telling the market he's not locked in, which keeps the narrative that SpaceX is building for its own empire, not becoming a glorified colo provider. Both things can be true if you squint, but only one reflects the actual strategy.
Key dynamics at play:
- AI labs are desperate for compute and will pay anything for guaranteed access
- Vertically integrated companies (SpaceX, Tesla, xAI) hold leverage because they can pull capacity whenever internal needs spike
- Public market investors want predictable revenue streams; Musk wants optionality
The implication for other AI infrastructure plays: if you're building data centers without your own AI workloads to backfill them, you're at the mercy of hyperscaler demand. If you're vertically integrated like Musk, you can have your cake (revenue from excess capacity) and eat it too (pull it back when you need it). That's why Meta is building its own clusters, why OpenAI is reportedly exploring chip design, and why Google never had to worry about this in the first place.
The broader pattern is clear. The companies winning the agent economy will be the ones who control their own compute stack. Anthropic is paying a premium for borrowed time. SpaceX is collecting rent on an asset it was going to build anyway. Guess who's in the better position when the lease expires.
The Implication
If you're investing in AI infrastructure, pay attention to who owns versus who rents. The vertically integrated players have structural advantages that won't show up in quarterly revenue but will show up in long-term positioning. Anthropic's $40 billion commitment looks like a massive win for SpaceX until you realize it's contingent on SpaceX not needing the capacity themselves.
For AI labs without their own compute, this is the warning shot. The big compute deals everyone celebrated in 2024 and 2025 were available because the vertically integrated companies hadn't scaled their internal workloads yet. That window is closing. Plan accordingly.