SpaceX has a CoreWeave problem that no one is talking about
After a no-good, very bad week for AI stocks, SpaceX gave the AI faithful—and all the Elon Musk acolytes—something to celebrate when it reported earnings on October 20. Its results beat analyst expectations across the board in almost every way, with revenue leaping 161% to $14.6 billion, and operating income hitting $2.8 billion. Even its so-called Super Intelligence segment—its AI business—reported $431 million in operating income, a welcome relief from the red ink that is a chief worry about AI companies. The stock surged 8%, JPMorgan doubled its price target to $500 a share, and Truist Securities called the results “the first data point during earnings season to support the bull case for AI cloud providers.”
What’s not to like?
Since I’m a curmudgeon and since you can read about what there is to like in a lot of other places, I’ll tell you what is not to like! But first, let me offer a trigger warning for those who can’t bear to hear anything bad about Elon Musk. Second, a caveat: This is meant to provide those who want to test their optimism a few counterpoints, not to reflect my complete views on the company.
To start with the obvious, SpaceX is priced less like an actual company than like a precious object that should be in a museum. It is now valued at $2.1 trillion, which is about 36 times its annualized revenue and roughly 55 times its annualized adjusted EBITDA (earnings before interest, taxes, depreciation, and amortization). That’s a premium that only a few other companies, like Palantir and Cloudflare, command. And as a source points out to me, that $2.1 trillion is almost twice the $1.28 trillion combined market value of the seven biggest defense giants—Lockheed Martin, General Dynamics, and so on. These are real companies, upon which the American economy depends. Does SpaceX’s premium reflect economic reality?
The profitability of both SpaceX overall and of the AI segment also merits a second look. The company’s accounting profit didn’t translate into free cash flow. Indeed, free cash flow burn was almost $10 billion, exceeding the expectations of a burn of $8.6 billion. In the AI business alone, SpaceX spent $16.4 billion in the quarter. There’s nothing wrong with spending cash in the expectation that the investments will turn into profits someday. But it does mean the company is dependent on the capital markets to supply the funding it needs to grow.
The larger issue is that SpaceX has effectively turned its AI business into a neocloud. Its own model, xAI, is badly lagging others, based on the reported token spend per model. XAI’s spend is a fraction of that of market leaders Anthropic and OpenAI. Whether because xAI was failing or because he spotted an opportunity, Musk pivoted to selling computing power to other players. That means that some of the same factors that affect a company like CoreWeave, which saw its stock plummet and the yield on its bonds soar last month because of its heavy cash burn and constant need to raise capital, also affect SpaceX’s AI business.
On the earnings call, Musk told analysts to expect 10 gigawatts of incremental compute to come online during 2027. According to a back-of-the-envelope calculation done by longtime skeptic Jim Chanos, for the most cutting-edge fully equipped data centers, that would require $500 billion-plus of capital expenditures, which is three times higher than the consensus view of what the entire company will spend in 2027. On the call, SpaceX management claimed they have debt lined up to finance the CapEx spree. “There are a number of available pools of debt capital that we are talking to at the moment relating to funding this incredible growth,” said CFO Brad Coffman. But it’s increasingly unclear if bond markets will absorb the sheer amount of debt all the AI companies plan to issue without charging higher rates—and, of course, higher rates make profitability more elusive.
Like other neoclouds, SpaceX also depends to some degree on customers that are themselves losing money. In the past quarter, SpaceX announced contracts with Anthropic, Google, and the startup Reflection AI, another AI model company. Google is obviously a rock-solid company. But Reflection AI loses money, as does Anthropic. With Anthropic’s IPO, the company will have a lot of cash with which to continue to pay suppliers like SpaceX. But someday, SpaceX’s customers have to be able to generate their own profits. Until that happens, even if SpaceX itself is producing profits, the capital markets still have to be willing to fund its customers, or SpaceX’s business goes away, too. In the short term, that could eventually pose a problem. “There does not appear to be a single enterprise customer for AI who pays the vendor more than it costs to deliver the service,” says longtime investor Roger McNamee.
In the most recent quarter, SpaceX’s revenue per megawatt of compute delivered—a common industry shorthand for measuring data center economics—soared from $8.5 million per megawatt in the second quarter to $19.5 million per average megawatt in this quarter, according to Chanos’s calculation. That number is better than the roughly $11 per megawatt that he says the average neocloud reports. But it’s below the $40 million to $50 million per megawatt that bulls argued last summer compute would fetch.
Which gets to the biggest issue. SpaceX is making a huge bet that the price of compute will continue to climb. Musk said that the company prefers to sell short-term contracts priced at the spot market, because he expects the price to keep rising. He isn’t hedging the risk that it will instead fall by locking in lots of long-term contracts. That’s probably why SpaceX’s revenue per megawatt is higher than other neoclouds’. “It is not impossible that we will have a six-month payback period for compute next year,” Musk said on the call.
But parsing the numbers shows just what kind of premium pricing Musk is counting on. SpaceX says it is on track for $100 billion in annualized recurring revenue in its AI business by year-end 2026, and is forecasting $500 billion to $1 trillion in AI annual recurring revenue by year-end 2027. If Chanos is right that the additional 10 gigawatts of compute capacity will require about $500 billion in capital expenditures, then SpaceX will have to produce between $1 and $2 of revenue for every dollar of capital expenditures. He estimates that currently, the AI industry is producing only about 30 cents of annual revenue for every dollar of infrastructure spending, so SpaceX is planning to get between three and seven times the industry’s current ratio.
There is certainly an argument that compute prices will only continue to rise, and that by positioning itself to take maximum advantage of the increases, SpaceX is making a smart bet. “The use cases are accelerating and we see it in the field every day, and the CapEx dollars will be in the trillions with a massive ripple impact across the economy,” writes one bull, Dan Ives of Yorkville.
But welcome to the great AI debate, in which for every bull there is a bear. So, of course, there’s also an argument that compute prices won’t rise. Anecdotal evidence suggests that the AI business is not all systems go. “Corporate America has been very uneven in its ability to extract value from its AI models,” Citadel CEO Ken Griffin said at Citadel’s Future of Global Markets conference in New York City. “I think all of corporate America is looking at their bills from Anthropic and OpenAI and going: ‘OK, we need to think about how to make sure we’re getting productivity out of this.’” Axios also recently reported that prices for tokens—a measure of AI usage—are plummeting. It’s possible that might stimulate more demand. But if corporate America does pull back on its spending on AI—if companies aren’t finding AI useful enough to continue to ramp up the spend—it’s hard to see how that doesn’t affect compute prices at some point.
Apart from the AI business, SpaceX also has its traditional space business, which is basically break-even, and Starlink. Starlink is a good, if not great, business—one that generated $2.4 billion of operating profit in the quarter. But combined, they are nowhere near enough to support a $2.1 trillion valuation. Before the IPO, Morningstar analyst Nicolas Owens argued that if investors were just getting the rockets and Starlink and AI didn’t work out, SpaceX could be worth as little as $40 a share.
SpaceX may well confound the skeptics, including me, as Musk has done so many times before. His companies have accomplished things people said were impossible. But it’s also worth remembering that not everything Musk says will happen does happen, at least not on the schedule he predicts, be it solar roofs, full self-driving cars, or Optimus robots. And at some point—who knows when, but there will be a point—the company will have to show that it is on a clear path to generate enough cash to validate the extraordinary prices that investors are paying.
Get Inc.’s free CPG newsletter, Packaged Goods, for the latest on the founders, investors, and trends reshaping consumer brands.
—By Bethany McLean, Chief Financial Correspondent
This article originally appeared on Fast Company’s sister website, Inc.com.
Inc. is the voice of the American entrepreneur. We inspire, inform, and document the most fascinating people in business: the risk-takers, the innovators, and the ultra-driven go-getters that represent the most dynamic force in the American economy.
内容来自 Fast Company,版权归原作者所有。
阅读原文 ↗