Space Exploration Technologies Corp. (NASDAQ: SPCX) is pursuing a massive expansion of its terrestrial AI computing capacity, aiming to increase power usage from 1.4 gigawatts to nearly 10 gigawatts by the end of 2027. According to a report by Financial Times Lex columnist John Foley, this expansion could require an investment of up to $500 billion, assuming each gigawatt of capacity costs roughly $50 billion to build.
The company is applying its rocket-building expertise to scale these data centers. SpaceX currently has agreements to rent computing capacity to Anthropic and Alphabet Inc. (NASDAQ: GOOGL) subsidiary Google. These contracts are valued at $1.25 billion and $920 million per month, respectively, representing approximately $26 billion in annualized payments once operating at full capacity. The company has also held talks with Microsoft Corp. (NASDAQ: MSFT) regarding a similar arrangement.
SpaceX CFO Bret Johnsen has stated that new computing investments can pay for themselves in less than a year. The company currently holds approximately $90 billion in cash, which Foley noted is far short of the estimated $500 billion cost. To support the build-out, SpaceX is reportedly seeking an additional $40 billion to purchase Nvidia Corp. (NASDAQ: NVDA) chips.
Foley acknowledged that the business model relies on renting capacity to customers who are in high demand. However, he noted potential risks, including the possibility of falling rental prices if competitors bring more capacity online and the ability of customers to terminate agreements with 90 days' notice after initial periods. Prediction markets currently give a 75% chance of at least 5,100 data centers being built on Earth by year-end, but only a 14% chance of a one-megawatt orbital data center going live before 2028.
