Key Points
- Meta Platforms and Anthropic are exploring a computing partnership valued at approximately $10 billion across a two-year period
- Discussions began when Anthropic presented the proposal in June; neither company has provided official statements
- Terms would involve monthly payment structures with flexibility for early termination by either side
- This agreement would be supplementary to Anthropic’s existing $45 billion SpaceX compute contract spanning three years
- Meta’s emerging cloud division, dubbed “Meta Compute,” comes as the company allocates $145 billion toward infrastructure in 2026
According to a July 17 New York Times report citing three individuals with knowledge of the matter, Meta Platforms and Anthropic are engaged in preliminary negotiations for a computing services agreement potentially reaching $10 billion across two years.
Sources indicate Anthropic initiated the proposal last June. The proposed framework calls for Anthropic to remit payments to Meta on a monthly basis throughout the two-year duration, with provisions allowing either company to terminate the partnership ahead of schedule. Representatives from both organizations have refrained from commenting.
CNN corroborated the ongoing negotiations, though a source cautioned that specific monetary amounts circulating in media reports may be conjectural. CNBC separately verified the discussions through its own channels.
Following the news, Meta’s shares experienced a decline of up to 6% during July 17 trading sessions, ultimately settling with a roughly 2% decrease at market close.
Strategic Implications for Both Parties
For Anthropic, securing adequate computing resources has emerged as a critical bottleneck throughout 2026. The organization has implemented restrictions on its premium offerings, including Claude Fable, as infrastructure availability struggles to match user demand.
The startup previously secured a $45 billion three-year computing arrangement with SpaceX this past May, providing access to the Colossus 1 facility located in Memphis. A partnership with Meta would represent an additional substantial source of graphics processing unit capacity.
Meanwhile, Anthropic is advancing toward an initial public offering, with Reuters indicating investment banks are organizing presentations to potential investors ahead of a potential October market debut. Establishing multiple computing partnerships prior to going public would bolster its investment narrative.
From Meta’s perspective, these negotiations represent the company’s first public step toward commercializing compute resources to external organizations. The initiative has been internally branded as “Meta Compute.”
CEO Mark Zuckerberg indicated in May that Meta was evaluating cloud computing services as a mechanism to demonstrate that its artificial intelligence investments can generate revenue streams beyond traditional advertising channels.
Meta’s Massive Infrastructure Investment
Meta projects capital expenditure reaching $145 billion for 2026, representing more than twice the $72 billion invested during 2025. The overwhelming majority targets AI computing hardware and datacenter expansion.
The social media giant reduced its workforce by 8,000 employees this past May while reallocating resources toward artificial intelligence infrastructure development. Additionally, the company recruited Dave Brown, formerly a high-ranking Amazon Web Services executive, suggesting cloud ambitions extending beyond individual transactions.
Meta currently leases computing capacity from various providers, including a $21 billion arrangement with CoreWeave and a $27 billion contract with Nebius.
What makes a potential Meta-Anthropic arrangement particularly noteworthy is that Meta develops its proprietary Llama language models, which directly compete with Anthropic’s Claude offerings. Such a compute partnership would position Meta simultaneously as both a rival and an infrastructure supplier to Anthropic.
However, this dual relationship isn’t unprecedented in the industry. SpaceX currently provides GPU resources to both Anthropic and Google. In today’s compute marketplace, organizations possessing available capacity are monetizing it by serving any customer requiring resources, irrespective of competitive dynamics.


