SpaceX's Google AI Pact Is Worth Up to $29 Billion, but Investors Shouldn't Bank on It

Starting this month, Alphabet (GOOGL -0.86%) (GOOG -0.83%) will fork over $920 million per month to Space Exploration Technologies (SPCX -3.28%) for access to compute capacity at one of its Colossus data centers. The contract, signed back in June, is one of several major deals SpaceX made for its AI infrastructure. Management expects to significantly ramp up its AI revenue as it builds new data centers.

But there are important terms built into SpaceX's contracts that investors shouldn't take lightly. And while Alphabet could pay a total of $29 billion over 32 months to Elon Musk's company, investors shouldn't bank on SpaceX receiving the full amount. Here's why.

Why did Alphabet sign a deal with SpaceX?

Alphabet runs a massive cloud computing business, but right now, it's extremely capacity-constrained. It's spending heavily to build new data centers and outfit them with new servers, with capex of around $200 billion this year and even more expected next year. Yet it simply cannot keep up with demand for compute on Google Cloud. Its cloud backlog grew to $514 billion last quarter, up from $460 billion in the prior quarter.

But Alphabet wants to continue growing its customer base and signing long-term cloud contracts. For now, that requires it to bolster its capacity by leasing some from third-party data center operators such as SpaceX, to secure contracts and provide immediate service.

During Alphabet's second-quarter earnings call, CFO Anat Ashkenazi called the use of third-party data centers a "bridging strategy while we build out more internal capacity." In other words, SpaceX and other neoclouds that are benefiting from Alphabet's shortage of compute today shouldn't expect it to remain a major customer in the long run.

SpaceX is potentially more susceptible to this risk than others contracting with Alphabet, as the terms of their deal allow either party to terminate it with 90 days' notice. That's not unique to the Alphabet contract: SpaceX includes that as a general term in all of its cloud contracts after an initial ramp-up period. That might enable SpaceX to sign customers more quickly and at higher-than-average prices, but it means that its long-term revenues are less reliable.

There's another reason Alphabet may have chosen SpaceX to provide additional compute capacity while it builds out its own: Alphabet owns a significant stake in SpaceX. As such, if its contract contributes to driving SpaceX's share price higher, that could in part compensate for the costs of using its data centers.

The post-IPO lockup period on Alphabet's SpaceX shares doesn't fully expire until the third quarter of 2027. It would be a surprise if Alphabet terminated its contract before then.

Supply-and-demand dynamics are driving everything

The big risk for SpaceX is that the current demand it's seeing for its AI business is driven entirely by economic forces, not by any competitive advantage its data centers offer over others. As the broad infrastructure build-out continues and supply and demand come into equilibrium, companies like Alphabet and Anthropic (SpaceX's biggest infrastructure customers) will be able to meet their needs with internal capacity. At that point, SpaceX could see a big drop in revenue.

During SpaceX's second-quarter earnings call, Musk estimated its ability to monetize its GPUs at $30 per watt to $50 per watt. That's extremely high pricing, and it's what underpinned CFO Bret Johnsen's assertion that it's seeing a payback period of less than one year on new compute capital deployment.

But if supply constraints ease, SpaceX won't be able to command that pricing anymore. That would decrease revenue and increase the payback period, significantly changing the calculus of data center capex.

SpaceX will have to prove that its AI servers offer some advantages over the competition in an environment where compute capacity isn't nearly as constrained. That could come in the form of lower costs, especially if it can successfully develop and scale its orbital data center plans.

Still, those cost advantages would have to be sufficient for it to offer a compelling deal to hyperscalers compared with owning all the servers themselves. Considering that this premise relies on SpaceX deploying at scale multiple technologies that it has yet to perfect, I see significant risk in investing in the company.

As things stand, SpaceX's AI revenue isn't durable, and the return on capital it actually achieves might be lower than it's expecting. With the stock already trading at a premium based on sales expectations, investors should remain cautious.