Social posts claim Anthropic pays xAI $1.25B/month to rent compute

Anthropic and xAI did publicly announce a compute partnership on May 6, but the most eye-catching number now circulating — $1.25 billion a month — did not appear in those original company posts. Anthropic said on its website that it had signed an agreement with SpaceX to use all of the compute capacity at Colossus 1, giving it more than 300 megawatts of new capacity and more than 220,000 Nvidia GPUs. xAI, in its own post, said Anthropic would get access to Colossus 1 for training, fine-tuning, inference and other workloads. The $1.25 billion figure appears to have come later from media reports that said SpaceX’s S-1 filing disclosed the payment terms. TechCrunch reported on May 20 that Anthropic would pay xAI $1.25 billion per month through May 2029, with a discounted rate for the first two months while capacity ramps up. Business Insider and other outlets published similar accounts a day later, also attributing the figure to the filing. (anthropic.com) ### Where did this claim actually start? An X post helped spread the number on social media, but the claim did not originate only on X. By May 20, multiple news outlets were already reporting the same figure and tying it to SpaceX’s IPO paperwork rather than to an unverified social post. That means the social chatter appears to be amplifying a filing-based report, not inventing it from scratch. (techcrunch.com) ### What have the companies themselves confirmed? Anthropic confirmed the partnership, the timing and the scale of the capacity, but not the monthly payment in its May 6 announcement. The company said the deal would bring more than 300 megawatts of capacity online within the month and directly improve service for Claude Pro and Claude Max subscribers. Anthropic also said the SpaceX deal sat alongside other compute arrangements with Amazon, Google, Broadcom, Microsoft, Nvidia and Fluidstack. (techcrunch.com) xAI confirmed the same partnership on May 6 and described Colossus 1 as having more than 220,000 Nvidia GPUs, including H100, H200 and GB200 systems. Its statement also said Anthropic had expressed interest in partnering on multiple gigawatts of orbital AI compute capacity. xAI’s post did not list price terms either. ### Why would Anthropic rent from a rival-linked system? (anthropic.com) CNBC reported on May 6 that Anthropic had agreed to use all of the compute capacity at the Memphis, Tennessee, site after Elon Musk merged SpaceX with xAI earlier this year. The arrangement is notable because Anthropic’s Claude competes with xAI’s Grok, and Musk had previously criticized Anthropic in public posts before praising its team the day the deal was announced. (x.ai) Anthropic itself framed the agreement as part of a broader scramble for capacity. Its May 6 post said it trains and runs Claude on AWS Trainium, Google TPUs and Nvidia GPUs, and that other large compute deals were already in place. That makes the xAI-linked Colossus deal part of a multi-supplier buildout, according to Anthropic’s own description. ### So what is verified, and what is still secondhand? (cnbc.com) The verified part is the existence of the partnership, the date, the Memphis Colossus 1 site, and the scale — more than 300 megawatts and over 220,000 GPUs. Those facts are in public statements from Anthropic, xAI and CNBC’s contemporaneous report. The less directly verified part is the exact payment schedule. (anthropic.com) Multiple outlets now say the contract is worth up to $1.25 billion per month through May 2029 and includes a 90-day termination clause, but that detail is being relayed from reports about a SpaceX S-1 rather than from the companies’ original press materials. The next concrete place to check is the underlying SpaceX filing itself or any follow-up statement from Anthropic, SpaceX or xAI. Until then, the monthly price should be described as reported in filing-based coverage, while the partnership itself is directly confirmed by both companies. (anthropic.com) (techcrunch.com)