'Clearly wrong': How Musk's Anthropic feud turned into an $84.5B compute deal

In September 2025, Elon Musk made a post on X that was pessimistic about Anthropic’s future and its chances of success. However, a year later he seems to have walked back on his dismissal of the AI company, as he says that he is clearly wrong.

This is coming as it became known that Anthropic is paying $1.25 billion a month to run its models on SpaceX hardware.

From “never” to “clearly wrong”

Musk’s 2025 post on the company behind Claude read, “Winning was never in the set of possible outcomes for Anthropic.” 

However, that view did not survive the compute deal. When users on X floated the idea that he could one day boot the lab off SpaceX’s servers to kneecap a competitor, Musk answered that he had been “clearly wrong about Anthropic.” 

He went further, stating that Anthropic is “obviously currently the leader in AI,” praised its Mythos/Fable model, and said abruptly cutting off a paying customer was “not my style.” 

As evidence, he pointed to Tesla’s 2014 decision to stop suing over its patents and to opening its Supercharger network to other carmakers.

However, that stance is not exactly out of altruistic pursuits, as Anthropic ranked among SpaceX’s largest customers as of July.

What Anthropic is actually paying

A SpaceX regulatory filing from May showed that Anthropic was handing it $1.25 billion every month, and that deal runs through May 2029 which is a run rate of $15 billion a year. 

It would likely be out of place for a seller to have dismissing takes for a buyer that is cutting checks the size mentioned in the disclosure. 

To put it into context, SpaceX’s total annual revenue sits near $18 billion, so a single AI tenant is worth close to a year of the rocket company’s sales. 

The arrangement gives Anthropic the entire 300-megawatt output of Colossus 1, a data center near Memphis, Tennessee, with more than 220,000 Nvidia GPUs. Musk’s xAI, which merged with SpaceX in February, owns that facility.

The Colossus 2 expansion and Claude’s limits

The relationship has since grown. Around the SpaceX IPO filing in May, Anthropic said it was extending past Colossus 1 into Colossus 2. “We’re expanding our partnership with SpaceX, and will be scaling up on (Nvidia) GB200 capacity in Colossus 2 throughout June,” Anthropic co-founder and chief compute officer Tom Brown wrote on X.

For users, the extra power translated into looser caps. Cryptopolitan reported that Anthropic doubled the five-hour rate limits on Claude Code for Pro, Max, Team, and enterprise seat plans, dropped peak-hour throttling for Pro and Max, and raised API limits for its Opus models. 

Anthropic also said it trains and runs Claude across AWS Trainium, Google TPUs, and Nvidia chips and has floated interest in multiple gigawatts of orbital compute with SpaceX.

Anthropic’s partners and its plan for the future

In a confidential IPO prospectus, Anthropic disclosed that it expects to spend a minimum of $518 billion over 10 years. The funds are spread across six partners, which include Google, Amazon, Microsoft, Broadcom and XAI. It stated that it is locked in to some commitments that it has to pay regardless of usage. 

However, Anthropic made it clear that it intends to build its own infrastructure with time to depend less on these partners who are also running their own respective AI labs.

For all the warm words from Elon Musk, the partnership is not locked in. Either side can walk away on 90 days’ notice. Google signed its own SpaceX compute deal, worth $920 million a month through June 2029, a sign the company plans to keep leasing capacity to rival AI labs. 

If you're reading this, you’re already ahead. Stay there with our newsletter.