Anthropic to pay xAI $1.25 billion per month for compute: an unusual deal between rivals
In the AI sector, business-partnership decisions beyond marketing narrative grow more interesting by the day. On 20 May 2026, according to TechCrunch, two major rivals in AI, Anthropic and xAI, signed a multi-year compute deal. Under the agreement, Anthropic will pay xAI $1.25 billion per month -- $15 billion per year; in return it will use the processing capacity of xAI's giant Memphis data centre, Colossus 2.
The deal reflects a new reality in the AI infrastructure market: all major companies are developing their own models and competing with one another as products, but the scale of the compute requirement often exceeds in-house capacity -- meaning sector rivals also have to maintain a customer relationship in 'capacity rental' terms. Anthropic flagged a 'December-February intensive compute' need for training the Claude Sonnet 4.6 and Opus 4.7 models; xAI, for its part, preferred to rent capacity for independent capital needs.
xAI's Colossus 2 data centre in Memphis, opened at full capacity in November 2025, holds 1.2 million Nvidia H200 GPUs, making it the world's largest single AI compute centre. xAI spent approximately $12 billion to build the facility; in addition, there is an earmarked budget of $8 billion to start Colossus 3 construction in September 2026. Anthropic's $1.25 billion monthly payment will be an important revenue source for repaying xAI's investment; estimates indicate Anthropic will cover approximately one-third of Colossus 2's annual operating budget.
Looked at from the financial side of Anthropic, the size of the deal is large but it is a 'fitting' line item within the financial structure. Anthropic's annual revenue in 2025 was about $8 billion; this year it is estimated to reach $18 billion. The $1.25 billion monthly payment to xAI will appear as a roughly 80% expense against the revenue line -- but thanks to Anthropic's containment of salary, sales and other operating costs, the company is aiming for its 'first profitable quarter' as of March 2026 (in line with a TechCrunch report on the same day).
Elon Musk has a direct history of rivalry with Anthropic; as an OpenAI co-founding member, his disputes with ChatGPT-developer Sam Altman are on the public agenda. For Musk, the xAI CEO, the business deal with Anthropic was directly positioned as a narrative of 'helping a rival instead of Sam Altman's OpenAI' -- Musk commented on Twitter, 'Anthropic going to us, not OpenAI, for compute shows the sector's confidence in my platform.' This statement further strengthens the strategic motivation of the deal.
Nvidia is directly affected by the deal. The compute agreement between Anthropic and xAI essentially constitutes a 'rental' of Nvidia GPUs -- the 1.2 million H200 chips in xAI's Colossus 2 are made by Nvidia. About 75% of the $1.25 billion monthly payment (~$940 million) goes to xAI, with the remainder spent on infrastructure operation, electricity and cooling costs. The long-term supply agreements xAI signed with Nvidia between 2023 and 2025 make the deal possible.
Digital-infrastructure analysts emphasise the symbolic importance of the deal for the 'maturation of the AI infrastructure market'. Previously the AI compute market was dominated by the AWS-Azure-Google Cloud trio. xAI's strategy of selling compute directly to customers in the 2024-2026 period weakens that trio's position. Major customers such as Anthropic can now rent capacity directly from other AI companies rather than 'renting cloud'; this points to a 'horizontal' structuring of the sector.
From the regulatory framework perspective, the deal is not yet at any formal review stage. The US Federal Trade Commission has voiced competition concerns over horizontal structuring in the AI sector on several occasions; but Anthropic and xAI continuing to be rivals while cooperating on the compute side does not reduce competition -- on the contrary, it enables both to develop their capabilities. The FTC's stance on such deals is unclear; analysts estimate a new guideline will be issued by the end of 2027.
The European Union's AI Act requires foundation-model developers such as Anthropic and xAI to submit annual transparency reports -- in those reports the source of compute capacity must also be disclosed. The details of the deal with xAI will pass through EU regulator review in the report Anthropic will publish at the end of 2026. But the technical specifics of the report data remain unclear; how deeply the AI Act's 'transparency' section will review will become clear in 2026-2027.
This article is in the nature of corporate-financial news and business analysis; it does not constitute or substitute for individual investment advice. Anthropic and xAI shares are not publicly available -- Anthropic has not announced any formal IPO plan, and xAI is privately held in Musk's group structure. Internal company developments do not generate share-price volatility, but they may affect other publicly listed companies in the AI sector (Nvidia, AMD, Microsoft, Google).