Meta in Talks to Lease $10 Billion in AI Compute to Rival Anthropic
News Summary
Meta Platforms and Anthropic are reportedly in early-stage talks over an arrangement that would see Meta lease as much as $10 billion worth of AI computing capacity to the Claude developer over roughly two years, according to multiple reports citing people familiar with the discussions. The negotiations, first reported in late July 2026 and detailed further in an August 27, 2026 New York Times report published at 5:00 a.m. Eastern Time, mark an unusual turn in a relationship where the two companies compete head-to-head for enterprise AI customers while potentially becoming business partners on the infrastructure side.
What the deal would involve
Under the terms being discussed, Anthropic would pay Meta in monthly installments for direct access to Meta's data centers and Nvidia GPU-powered infrastructure. The arrangement remains preliminary, and terms are subject to change, but people briefed on the talks describe it as one of the largest compute-leasing agreements struck outside the traditional cloud providers such as Amazon Web Services, Microsoft Azure, and Google Cloud. No final agreement has been signed as of publication, and both companies have declined to comment publicly on specifics.
Why Meta has spare capacity to sell
The talks come as Meta sharply raises its own capital spending. The company's 2026 capital expenditure guidance, updated in April 2026, now stands at $125 billion to $145 billion, more than double the roughly $72 billion it invested in 2025. Much of that spending has gone toward data centers and Nvidia chips to support Meta's Superintelligence Labs and its family of Llama-based AI products. Meta is the only one of the four major U.S. hyperscalers without an existing cloud infrastructure business, and executives have said the company routinely fields inbound requests from outside firms asking to buy compute at a premium.
Speaking at Meta's annual shareholder meeting in May 2026, CEO Mark Zuckerberg said a formal cloud computing business was "definitely on the table," adding that "almost every week there are different companies that come to us from outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium." Meta shares rose roughly 9 percent in early July 2026 after the company signaled it would begin selling excess AI computing capacity to outside customers, with analysts noting the move could offset investor concerns about elevated capital spending, even if it comes with thinner margins than Meta's advertising business.
Why Anthropic needs the capacity
For Anthropic, the talks reflect an ongoing compute shortage rather than a shift in strategy. The company has reported explosive growth in 2026, with CEO Dario Amodei citing roughly 80-times annualized growth in its Claude Code product during the first quarter and a revenue run rate reaching about $30 billion. That growth has repeatedly outpaced available graphics processing unit capacity. In late March 2026, Anthropic tightened usage limits on Claude during weekday peak hours, and the company has said access to chips remains its central bottleneck even as its valuation approached $1 trillion in secondary-market trading in May 2026.
To close the gap, Anthropic has already lined up large infrastructure commitments elsewhere, including a reported $40 billion arrangement with Google and a $25 billion arrangement with Amazon, both of which are structured to deliver additional compute in stages through late 2026 and into 2027. A potential Meta deal would add a further, faster source of capacity, since it would draw on infrastructure Meta has already built rather than requiring new construction.
A complicated competitive backdrop
Meta and Anthropic are simultaneously rivals in the market for enterprise AI tools and developer services. In August 2026, Meta introduced its first AI coding agent, positioned as a direct competitor to coding-focused products from Anthropic and OpenAI. At the same time, Meta's own Superintelligence Labs has drawn scrutiny over internal strategy disagreements, and the broader AI industry has seen public friction between rival labs, including a widely covered advertising dispute between OpenAI and Anthropic executives earlier in 2026.
Analysts describe the potential Meta-Anthropic arrangement as evidence that the AI industry is increasingly separating two distinct businesses: building and selling AI models, and operating the physical infrastructure that powers them. Under that view, a company can be a customer's competitor in one market while acting as its landlord in another. If finalized, the deal would also give Meta a concrete answer to investors who have questioned whether its enormous data center spending can generate revenue beyond improving its own consumer products.
What happens next
Because the discussions remain in early stages, terms including price, duration, and the exact scope of data-center access could still change substantially, and there is no guarantee the two companies will reach a final agreement. Both Meta and Anthropic have continued to expand infrastructure commitments with multiple partners in parallel, suggesting that even if this particular deal does not close, the broader trend of AI labs leasing compute from one another's parent companies is likely to continue as chip and power constraints persist industry-wide.