SpaceX (NASDAQ:SPCX) CEO Elon Musk says artificial intelligence will overtake everything else the company sells as soon as next month.
In an all-hands meeting, Musk said AI revenue “will exceed all other SpaceX revenue probably in September, like next month,” and will “significantly exceed” the rest of the business in the fourth quarter.
The forecast implies a rapid shift. AI generated $2.56 billion of the $7.81 billion SpaceX reported for the second quarter, behind the $4.29 billion from its Starlink-led connectivity unit.
The company is targeting a roughly $100 billion annualized revenue run rate by year-end, more than triple the roughly $31 billion pace implied by the second quarter.
The Math Behind the $100 Billion Path
AI and semiconductor research firm SemiAnalysis estimates that one megawatt of cutting-edge Nvidia Corp. (NASDAQ:NVDA) compute running frontier AI inference can support roughly $100 million in annual revenue.
Inference is the computing used each time a trained AI model responds to a user.
Because those economics are so lucrative, the firm argues scarce, immediately available compute could rent for around $30 million to $50 million per megawatt each year.
SpaceX expects to have more than two gigawatts of compute capacity online by year-end. At SemiAnalysis’s $40 million-per-megawatt assumption, two gigawatts of fully rented capacity would imply $80 billion in annualized revenue.
That is illustrative math based on SemiAnalysis’s assumptions rather than a SpaceX forecast, and the company may keep some capacity in-house to train Grok rather than rent it out.
SpaceX has separately said its compute investments are paying back quickly. CFO Bret Johnsen said new AI investments have payback periods of less than a year.
Why SpaceX Could Build Faster Than Rivals
Musk has argued that building data centers is relatively simple compared with SpaceX’s day job, saying the work “ain’t rocket science” and joking that rockets “desperately want to blow themselves into tiny pieces.”
The record offers some support. Musk’s xAI built Colossus, a 100,000-GPU supercomputer, in 122 days, then doubled it to 200,000 GPUs in another 92 days. Nvidia said training began just 19 days after the first rack arrived.
SemiAnalysis says Colossus 2 is repeating that playbook at far larger scale, and estimates peak construction labor per gigawatt at roughly one-third the level of even the fastest conventional developers.
SpaceX Is Selling Speed
A recent deal with Alphabet Inc. (NASDAQ:GOOGL) shows what that speed is worth.
SemiAnalysis estimates SpaceX is effectively charging Google about $14 per hour for GB300 capacity, versus roughly $3 per hour for comparable conventional compute.
The premium reflects timing. Conventional data centers can take 12 to 18 months to come online, while SpaceX can deliver large blocks of capacity within months.
Prediction-market traders are leaning bullish in the near term, with Polymarket pricing roughly a 73% chance SpaceX shares reclaim $140 by the end of August.
The next earnings report should show whether Musk’s prediction is translating into the kind of AI revenue growth needed to support the $100 billion target.
Image: Shutterstock
Login to comment