SpaceX's Stock Is an Uptime Bet, and Its Data-Center Overhaul Is the Fix

On the morning of September 3, Grok went dark. xAI's chatbot stopped responding on X and across its apps for about three and a half hours, and the company's apology named the cause in plain terms: an "outage at our Memphis compute center." The phrasing that mattered came next — the note was addressed not just to users but to "impacted compute partners," the outside customers who rent the machines in that building.

That is the sentence that explains the news cycle around SpaceX's (SPCX) stock. The headline reads like a warning: the company is overhauling its data-center strategy, pulling rocket and Starlink executives into the AI group after reliability failures in Tennessee and Mississippi. Slower build-out, delayed sites in Texas. In the ordinary logic of growth stocks, that should be bad.

Look at what SpaceX actually is now, and it flips.

The largest IPO in history was not really a rocket story

SpaceX priced the biggest public listing ever in June — worth about $1.77 trillion at its $135 offer price, raising a record $85.7 billion once underwriters exercised their extra allotment. It then traded up to a peak near $225 before giving much of it back; shares sit around $140 today.

Read the financials and the label stops fitting. In the company's first quarter as a public firm, total revenue rose 92% year over year to $7.8 billion. The AI business contributed $2.6 billion of that, up 247%, and it swallowed $15.8 billion of capital spending in a single quarter — about 86% of everything the company spent. The rockets and Starlink are real and profitable. Starlink alone made a $4.4 billion operating profit last year. But the thing the market is paying a trillion-plus dollars for is the AI-compute business layered on top of them.

That business is a landlord. xAI doesn't just run models; it rents out gigantic clusters of GPU compute to hyperscalers. Anthropic is contracted to pay roughly $1.25 billion a month through 2029. Google signed a compute deal worth as much as $2.17 billion a month. The revenue is contracted, recurring, and enormous — and it is all hostage to one thing: the machines staying on.

When you sell uptime, downtime is the product failing

That is the layer the overhaul is really about. During SpaceX's aggressive early build-out, some facilities in the Memphis/Southaven cluster ran for months without backup cooling or power, assembled with temporary gear — portable chillers, mobile gas turbines, Tesla storage packs — that could not hold a 99.9% uptime target. Models got interrupted mid-training. And on September 3, the biggest customer-facing symptom surfaced: a 3.5-hour outage at the Memphis compute center that hit Grok and, by the company's own admission, third-party compute customers too.

Here is the disconnect between the headline and the business. For a hyperscaler deciding whether to trust a landlord with billions in monthly commitments, a data center that blinks is a data center that loses the contract. A company that slows down to become reliable isn't shrinking its growth engine — it's protecting the revenue it hasn't earned yet. That's why the stock actually gained about 6% on the day the overhaul plans surfaced, to settle near $149. The market read the shake-up as adults taking over a wall being built too fast, not as retreat.

The number that has to work

So the bullish read is coherent: SpaceX has compounding rocket and Starlink cash, a missionary founder, and a bet on the largest compute market in history. The asymmetry is real, and it's why this reset matters.

But the same data that supports the story sets the bar for it, and the gap is the honest risk. The AI segment reported a positive $1.1 billion in adjusted profit last quarter — a number that deliberately strips out the depreciation of the very machines the whole valuation rests on. Under plain accounting, the segment still lost $1.3 billion operating. The bull case fast-forwards from 1.4 gigawatts of capacity at the end of June to 10 gigawatts by the end of 2027, a roughly sevenfold step-up that Elon Musk himself projected, and it turns today's $2.6 billion of quarterly AI revenue into a projection near $322 billion annually that helped justify that $1.77 trillion IPO price.

The reliable build-out makes that trajectory more believable, not less — a data center that holds uptime can hold a hyperscaler contract, and a hyperscaler contract is the compounding asset. The catch is that the stock is still priced as if all of it lands on schedule, with the customer concentration as the single point of failure: one or two rental deals with termination clauses sitting atop one enormous physical campus. The reset removes the operational excuse for missing those numbers. It doesn't remove the arithmetic.

What changed this month isn't that SpaceX's AI growth is over. It's that the company stopped pretending reliability and speed are the same thing, at roughly $140, after the market had already marked the stock down a third from its high. For a buyer, the question was always whether the compute would actually stay lit. This week, the people running it started acting like the answer depends on the job, not the sprint.