Elon Musk’s xAI is no longer just a chatbot fight with OpenAI and Anthropic. SpaceX’s Q2 earnings show the business is leaning hard into a different identity: a compute landlord, with capital spending on AI exploding from Q1 to Q2 and consuming 82.7% of SpaceX’s 2026 capex. That shift is happening even as Grok’s consumer momentum weakens, leaving Musk’s AI arm increasingly tied to data centers, GPU rental revenue and a wager that infrastructure can scale faster than a chatbot brand.
The numbers are blunt. xAI’s AI capital expenditures more than doubled from $7.7 billion in Q1 to $15.8 billion in Q2 2026. Over the full 2026 capital budget, xAI spend reached $23.5 billion of SpaceX’s $28.4 billion total. SpaceX also said it spent more on AI in Q2 than it did in all of 2025, when xAI capex totaled $12.7 billion. For a company that has sold itself around rockets, satellites and orbital ambition, the near-term profit story is now about servers, not software.
The clearest sign of the pivot is in earnings math, not product hype. xAI swung on an adjusted EBITDA basis from a loss of $609 million in Q1 to a gain of $1.1 billion in Q2, and PitchBook said nearly all of that growth came from renting out access to GPUs. In other words, the business is not yet being carried by consumer adoption of Grok. It is being carried by customers paying for the infrastructure underneath it.
That shift lines up with what SpaceX CFO Bret Johnsen told investors on the Q2 call. “On the AI compute side, we’re able to deploy capital in such a way that we’re getting less than a one-year payback,” he said. That comment matters because it explains why the company is willing to pour money into data centers at a scale that would look extreme almost anywhere else. The payback period, at least according to management, is short enough to justify the spending surge.
The scale of the buildout is already visible in the customer contracts embedded in xAI’s filings. TechCrunch, citing the SpaceX S-1, said Anthropic signed a deal to pay $1.25 billion per month through May 2029 for the full 300MW Colossus I capacity. Another filing detail, reported by otontechnology.com, showed Google signed a separate contract at $920 million a month for 32 months, a commitment of about $29.4 billion. Those numbers point to a business model that is drifting away from a pure model-versus-model contest and toward a rent-the-rack race.
The market problem for Musk is that the product story is not keeping pace with the infrastructure story. The evidence pack says Grok downloads fell more than 50% from a January peak of 20 million. That does not make Grok irrelevant, but it does suggest the consumer app is not yet the breakout engine that would explain the company’s strategic posture on its own. For now, the demand signal appears to be coming from compute customers, not from users rushing to chat with Musk’s model.
That is why investors and analysts are framing xAI less as an app company and more as a neocloud. Nicolas Owens, an equity analyst covering SpaceX for Morningstar, put it bluntly: “Frankly, the only way for them to make money right now is in data centers, and they’ve been building it very fast.” He added: “It’s still really early days for this AI business. I think it’s going to be a decade before they turn a profit on this because they’re building as fast as they can and spending hand over fist for data centers.” The tone is cautious, but the message is clear: the business is being valued more for infrastructure momentum than for chatbot share.
Owens also pointed to the physical footprint behind the spending, including Colossus II in Tennessee and Mississippi and the expansion of Colossus I in Memphis. The implication is that xAI’s economics now look like a classic infrastructure buildout: heavy upfront capital, followed by the hope of long-lived contracted revenue. That is a much different profile than a software startup chasing downloads and subscriptions.
Not everyone sees the pivot as retreat. Tejas Dessai, Global X director of thematic research, said: “I wouldn’t call it a stepping back. I would call it doubling down on what is working for them.” That view reflects a broader investor logic: if model competition is brutal, then owning the compute stack may still be a better place to make money than fighting for consumer mindshare against OpenAI and Anthropic.
SpaceX investors appear willing to accept that trade-off, even if they are uneasy about Grok’s competitive standing. Michael Mealling, GP at Starbridge Venture Capital, which is an investor in SpaceX, said: “It’s an area SpaceX has never gone into, and it’s the one I am the least confident in.” He added: “I just don’t see the adoption rate on Grok that would suggest it’s going to unseat Claude and some of the bigger models, even some of the Chinese models that are coming out.” His skepticism is directed at the product layer, not the data center layer, which is exactly where the company now seems to be concentrating.
Mealling argued that SpaceX may have an edge in the hard parts of data center construction because of its experience with space engineering. He cited heat dispersal, power distribution and fire management as problems both industries have to solve. “There’s a tremendous amount of capex and opex when it comes to data center operations that can be pulled out if you come at it from the mindset SpaceX does,” he said. “A huge amount of this cross-interdisciplinary stuff can be done. So many of these issues the space sector has been dealing with since the 1950s.” Whether that proves out operationally is still an open question, but the logic helps explain why investors are not treating xAI’s spending as pure waste.
Even if the payback math is real, the spending pace is staggering. PitchBook said xAI’s AI capital expenditures more than doubled in a single quarter, and that the business spent more on AI in Q2 than it did in all of 2025. SpaceX’s 2026 capex mix shows just how concentrated the bet has become. xAI accounting for 82.7% of total capex is not a side project. It is the project.
SpaceX appears prepared to keep the machine running. Yahoo Finance, citing the earnings call, said the company expects capex to remain at roughly current levels over the next two quarters, Q3 and Q4 2026. SpaceX also said in its S-1 that it “expects to enter into additional similar services contracts” beyond Anthropic and Google. That means the revenue side may keep growing, but so may the bill.
For now, Musk’s AI story is less about whether Grok can catch ChatGPT or Claude and more about whether xAI can turn a sudden flood of demand for compute into a durable business. The latest numbers say the answer may already be yes, at least for the servers. The harder question is whether the model layer ever catches up.
