Zoom turned $100 million into $1 billion while Anthropic commits to spending $45 billion on compute from Musk's SpaceX—a single customer deal worth more than most tech IPOs.
The Summary
- Zoom's early 2023 bet on Anthropic has generated a $1 billion gain, turning the videoconferencing company into an accidental AI investing winner
- Anthropic will pay SpaceX $1.25 billion per month through May 2029 for access to compute capacity at its Colossus and Colossus 2 data centers in Tennessee
- Either party can terminate the deal with 90 days' notice, but if it holds, SpaceX earns over $40 billion from a single customer
- The compute powers inference for Claude's growing user base, while SpaceX monetizes unused capacity without sacrificing its own needs
The Signal
The numbers reveal two stories about the AI economy's current state. Zoom invested in Anthropic in early 2023 when most enterprise software companies were still debating whether AI agents were vaporware. That stake is now worth $1 billion more than they paid. Meanwhile, Anthropic is spending at a pace that makes even OpenAI's compute bills look modest.
The $1.25 billion monthly payment to SpaceX translates to $15 billion annually. For context, that's roughly what Netflix spends on content in a year, except Anthropic is buying raw computational horsepower to run Claude inference at scale.
"This structure allows us to monetize unused compute capacity in our infrastructure, while still permitting reallocation of the capacity for our own inter[nal needs]."
The timing matters. SpaceX's S-1 filing, released for its upcoming IPO, disclosed the full scope of the Anthropic deal. SpaceX isn't just launching rockets anymore. It's operating two massive data centers—Colossus and the newer Colossus 2—and renting them to one of the hottest AI labs in the world. The company expects to sign similar deals going forward, positioning itself as critical infrastructure for the agent economy.
Anthropic's compute chief Tom Brown confirmed the compute powers inference, not training. That distinction is key. Training models is a one-time capital expense. Inference is the ongoing cost of actually running AI systems for users. As Claude usage grows, so does the compute bill. Anthropic clearly believes its revenue trajectory justifies this burn rate.
Key deal mechanics:
- Anthropic gets discounted rates for May and June 2026
- Standard $1.25 billion monthly payments kick in after the discount period
- 90-day termination clause gives both parties an exit
The wildcard is the termination clause. Either side can walk away with 90 days' notice. If Anthropic's growth slows or if it finds cheaper compute elsewhere, SpaceX loses its biggest data center customer. If SpaceX needs the capacity for its own projects or gets a better offer, Anthropic has to scramble for alternatives. That flexibility cuts both ways.
The Implication
Watch Zoom's next moves. A $1 billion paper gain on an AI bet suggests they saw something in 2023 that most didn't. The question is whether they double down or cash out. If they're holding, it's a signal they think Anthropic's valuation has further to run.
For Anthropic, the SpaceX deal is both a strength and a vulnerability. Spending $45 billion over three years means they're extremely confident in Claude's revenue growth. But that 90-day termination window means they can't get comfortable. If usage spikes beyond projections, they'll need even more compute. If it disappoints, they're locked into billion-dollar monthly payments they may not need. SpaceX just became the most important company in AI that nobody talks about as an AI company.