IT之家 AISource publication

SpaceX's Q3 AI (SI) spending hits a new high, with related business revenue of 7.6 billion USD surpassing Starlink for the first time

IT Home reported on October 21 that SpaceX's second earnings report released today (the 21st) showsthe company's AI (listed as SI in the official announcement; referred to as AI below to fit Chinese conventions) spending has hit a new high, and capital expenditures may even double soon.

IT Home reported on October 21 that SpaceX's second earnings report released today (the 21st) showsthe company's AI (listed as SI in the official announcement; referred to as AI below to fit Chinese conventions) spending has hit a new high, and capital expenditures may even double soon.

SpaceX's AI capital expenditures reached 16.4 billion USD in the third quarter of this year (IT Home note: approximately 109,738 million RMB at current exchange rates), continuing the trend of increasing investment over recent quarters.

SpaceX continues to expand infrastructure construction for its AI businesses such as Grok; in the third quarter,total capital expenditures were approximately 19.6 billion USD(approximately 131,150 million RMB at current exchange rates), roughly in line with Wall Street analysts' expectations.

Of this,16.4 billion USD (approximately 109,738 million RMB at current exchange rates) went to the AI business, slightly below analysts' expectation of 16.9 billion USD (approximately 113,083 million RMB at current exchange rates).

SpaceX expects capital expenditures to increase substantially this quarter. The company's CFO Bret Johnsen predicts that as more computing facilities come online, capital expenditureswill rise to between US$35 billion and US$40 billion(approximately RMB 234.196 billion to RMB 267.653 billion at current exchange rates). He also expects the new investments to pay off quickly: "At current pricing, newly deployed compute pays for itself in less than a year."

After the earnings report was released, SpaceX shares rose 4% in after-hours trading.

SpaceX CEO Elon Musk pointed out on the earnings call that the biggest bottleneck for AI infrastructure construction isno longer chips: "As for SpaceX's superintelligence business SpaceXSI — 'sexy' or whatever, whatever. I think people are starting to realize that the main constraint on deploying superintelligence is not chips, it'spower supply, cooling, and infrastructure。”

SpaceX's Q3 AI business revenue grew sharply quarter-over-quarter,reaching 7.6 billion USD(approximately 50,854 million RMB at current exchange rates), exceeding analysts' expectation of 6.3 billion USD (approximately 42,155 million RMB at current exchange rates).

With revenue of 7.6 billion USD (approximately 50,854 million RMB at current exchange rates),the AI business has officially surpassed SpaceX's Starlink business. That quarter, Starlink's revenue was 5.2 billion USD (approximately 34,795 million RMB at current exchange rates).

According to Business Insider, SpaceXSI's computing power rental business has achieved some success, with clients including Anthropic, Google, and Reflection AI. On the previous earnings call, SpaceX CFO Bret Johnsen disclosed that computing rental agreements brought the company an additional 1.6 billion USD in revenue (approximately 10,706 million RMB at current exchange rates).

Musk said SpaceX is receiving computing rental requestsmore than it can handle: "Honestly, all the hyperscalers, plus dozens of smaller companies, keep pestering us to rent more computing power. If we agreed to all of them, our existing computing power wouldn't be nearly enough to rent out — it could be rented out several times over."

As for how SpaceX's computing power will be allocated, Musk responded that the specific ratio will be adjusted month by month: "Roughly speaking, we want a 50-50 split, half reserved for our own use, and the other half rented out to other companies. But the ratio could also become 60-40, with either side taking the 60%."

Original source

IT之家 AI

Content notes

Original publication and rights belong to the source.

Machine translation · Refer to the original