Musk's $15.8B AI capex defense eases memory-chip oversupply fears

Elon Musk used SpaceX's first earnings call as a public company to push back on concerns that AI infrastructure spending is unsustainable, arguing that data centers are a "trivial problem" compared with building reusable rockets.

"Data centers are a trivial problem compared to making reusable rockets," Musk said on the call Tuesday, adding that applying even a fraction of SpaceX's aerospace engineering discipline to AI infrastructure produces "an amazing outcome."

SpaceX reported second-quarter revenue of $7.8 billion, up 92 percent year over year, with AI revenue of $2.6 billion — up 247 percent — while AI capital spending reached $15.8 billion, or 86 percent of total capex of $18.4 billion. The company signed $14.1 billion in contracted cloud services agreements, including deals with Anthropic and Google.

The comments carry weight beyond SpaceX. Semiconductor and memory-chip stocks have been under pressure as investors question whether AI-driven demand can absorb the massive buildout of data centers and GPU capacity. Musk's assertion that SpaceX will receive "a very significant percentage" of Nvidia's GPUs next year — and its target of 20 gigawatts of compute capacity by end-2027 — could provide a floor for those concerns.

The payback math

SpaceX Chief Financial Officer Bret Johnsen said the economics of AI infrastructure investments are improving rapidly, with a less than one-year payback on new capital deployments for compute. That contrasts with traditional data center investments, which typically take years to recover upfront costs. Johnsen added that SpaceX signed another $6.7 billion in cloud computing contracts since the end of the second quarter and was on track to reach a $100 billion annualized revenue run rate by year-end.

The company ended the quarter with 1.4 gigawatts of nameplate compute capacity, up from 1 gigawatt in the first quarter and 0.4 gigawatts a year earlier. Musk said SpaceX's "tentative target" is 20 gigawatts of power and cooling capacity online by end-2027, with a likely floor of 15 gigawatts even with delays.

Nvidia's role and investor skepticism

Musk said SpaceX will build its AI computing platform exclusively on Nvidia's Vera Rubin architecture, calling it the industry's best AI computing system. "Our understanding with Nvidia is that we will receive a very significant percentage of their GPUs next year," he said. The commitment shows the deepening relationship between SpaceX and Nvidia, which also supplies GPUs to hyperscalers including Microsoft, Amazon, and Alphabet's Google.

Despite the bullish tone, SpaceX shares fell 9 percent in after-hours trading Tuesday, dropping well below the $135 IPO price from June. Short sellers hold about 206 million shares, roughly 32 percent of the publicly tradable float, according to S3 Partners. Musk warned on X that institutions maintaining large short positions for a long time have an "extremely low survival probability."

"I believe the numbers. I would say that yes, those numbers are aggressive, but it's not a fantasy. The pieces exist, they just require flawless execution," David Wagner, portfolio manager at Aptus Capital Advisors, said.

The tension is straightforward: AI revenue is growing faster than expected, but the company is still spending far more than it earns. The AI division recorded an operating loss of about $1.26 billion in the quarter, and depreciation and amortization reached $2.85 billion — of which AI accounted for $1.89 billion. The more SpaceX builds today, the higher the depreciation, maintenance, energy, and financing costs in future profit statements.

For semiconductor investors, the key question is whether customers like Anthropic and Google can absorb the newly added compute capacity over the long term, and whether AI infrastructure revenue can continue to grow faster than capital expenditure. Musk's confidence — and his willingness to commit to Nvidia's Vera Rubin architecture — suggests the AI compute buildout has no near-term ceiling. But the market's reaction to SpaceX's own numbers shows that even the most aggressive AI capex plans face scrutiny when the losses are visible.

This article is for informational purposes only and does not constitute investment advice.