Elon Musk Mocks Anthropic With Chilling AI Warning: Company Could Create the Very Superintelligence It Fears

Anthropic has agreed to pay Elon Musk's SpaceX $1.25 billion (£940.02 million) a month for computing power, even as Musk mocks the AI company's fears about superintelligence. The arrangement places SpaceX at the centre of a debate over the risks of the technology it is helping Anthropic develop.

Bloomberg identified Anthropic as SpaceX's largest single customer in the third quarter of 2026, reporting that it accounted for approximately 34.5 per cent of the company's total revenue.

Musk's Anthropic Jab Fuels AI Safety Debate

The exchange began when X user James Douma suggested that Anthropic could end up building the very superintelligence it fears.

Musk responded: 'The most entertaining outcome is the most likely, especially if ironic.'

His remark was a sarcastic response to Douma's hypothetical, rather than an explicit prediction. It has drawn attention to Anthropic's position in the debate, as chief executive Dario Amodei has repeatedly highlighted the potential dangers of increasingly powerful AI.

SpaceX's $1.25 Billion Monthly Anthropic Agreement

SpaceX's May 2026 cloud services agreements give Anthropic access to computing infrastructure at its Colossus and Colossus II facilities in Memphis, Tennessee. The company's regulatory filing describes approximately 325,000 Nvidia GPUs, alongside processors, storage and high-speed networking.

Anthropic agreed to pay $1.25 billion monthly through May 2029, with reduced fees during the May and June capacity ramp-up.

The agreement includes an initial three-month period followed by 90 days' termination notice, creating a six-month minimum commitment. Musk previously described the arrangement as a 180-day lease, saying SpaceX wanted the flexibility to reclaim capacity if its own demand increased.

The financial significance of the deal is underlined by Bloomberg's report that Anthropic accounted for approximately 34.5 per cent of SpaceX's third-quarter revenue. Bloomberg identified the customer as the unnamed company in SpaceX's filing.

SpaceX's AI segment also reported a $2.5 billion (£1.8 billion) operating loss in the first quarter of 2026, against $818 million (£615 million) in revenue. The figures highlight the scale of the segment's costs and the importance of customers buying computing capacity.

SpaceX Backs Anthropic While Building Its Own AI

The deal puts SpaceX in the position of supplying computing capacity to Anthropic while pursuing its own AI ambitions. Its filing said Grok-5 was being trained at Colossus II.

SpaceX said it had enough capacity to support its own AI requirements while meeting third-party agreements. Anthropic, meanwhile, gains access to infrastructure for developing and operating its Claude models. The arrangement links the two companies commercially, even as they pursue their own AI projects.

AI Acceleration Puts Washington Under Pressure

A November 2026 assessment by AI safety organisation METR found that Anthropic's models had improved on benchmark tasks approximately 1.9 times faster since October 2025. The estimate carries substantial uncertainty.

Researchers found no evidence of highly significant, fully autonomous contributions to the most important stages of model training.

Anthropic has also given six METR employees access to internal tools and systems for independent evaluation. Specified confidential information can be redacted.

In Washington, Reuters reported that Democrats plan to pursue independent AI testing and corporate liability legislation when the new Congress convenes in January 2027. Lawmakers including Chuck Schumer, Lori Trahan and Greg Casar have backed different aspects of the proposed regulatory push.

Anthropic has separately pitched a draft bill proposing government intervention in narrowly defined, imminent high-risk scenarios. The proposal remains under discussion and has not become law.

Musk's comment contrasts with SpaceX's commercial relationship with Anthropic: the company is supplying infrastructure to a customer whose concerns about AI risks he has mocked. The deal illustrates how companies competing in AI can also depend on one another for computing capacity.