21 hours ago
10
Meta is in early discussions to lease computing capacity to Anthropic under a two-year agreement worth up to $10 billion, according to The New York Times. The proposed arrangement would give Anthropic access to computing infrastructure supplied by Meta while creating a potential external revenue stream for the company.
The reports did not identify the facilities, chip models, or amount of capacity involved. Anthropic would reportedly pay in monthly instalments, while both companies would retain the option to terminate the agreement early.
Neither company commented on the reported negotiations. The discussions remain preliminary, and the reported $10 billion figure represents the maximum potential value rather than a confirmed contract amount.
Anthropic broadens its infrastructure mix
Anthropic has signed several large infrastructure agreements as it expands the computing capacity supporting its models and services. A Meta agreement would add to its existing relationships with Amazon Web Services, Google, SpaceX, and specialist data centre operator TeraWulf.
Anthropic already operates Claude across several accelerator platforms, including AWS Trainium, Google TPUs, and Nvidia GPUs. The company has also secured multiple gigawatts of Google and Broadcom TPU capacity scheduled to begin coming online in 2027.
Anthropic agreed to pay SpaceX about $1.25 billion a month through May 2029 for access to its Colossus computing infrastructure. It has also entered a 20-year data centre lease with TeraWulf expected to generate about $19 billion in contracted revenue over the initial term.
The agreements differ in structure and technical scope. Anthropic’s work with AWS includes long-term Trainium capacity and engineering collaboration with AWS’s Annapurna Labs on processor optimisation and the Neuron software stack.
The available reports do not specify whether Meta would provide similar software integration, dedicated computing clusters, or another form of infrastructure access. The companies have also not disclosed how Anthropic’s workloads would be deployed or managed under the proposed arrangement.
Meta looks beyond internal workloads
A deal would mark a change in how Meta uses its data centre estate. The company has mainly built infrastructure and acquired computing capacity to support its own AI development, advertising systems, and consumer services.
Meta expects capital expenditure of between $125 billion and $145 billion in 2026, largely reflecting spending on data centres and other AI infrastructure. That compares with capital expenditure of about $72 billion in 2025.
Meta operates a global data centre network using CPUs, GPUs, and internally developed MTIA accelerators. The company is also expanding its Richland Parish campus in Louisiana to 5 gigawatts of computing capacity.
Meta has not identified which facilities, if any, would support the proposed Anthropic agreement. It also continues to obtain computing capacity from external infrastructure providers, including CoreWeave and Nebius.
Meta has continued to acquire external capacity while building its own data centres. The company has not identified which part of its infrastructure estate would be involved in the Anthropic proposal.
Chief executive Mark Zuckerberg has said that companies regularly approach Meta about purchasing computing capacity at a premium. At a shareholder meeting in May, he said entering the cloud-computing market was under consideration and that Meta could supply external customers if it built more infrastructure than it required internally.
Meta has organised part of its infrastructure expansion under an internal initiative referred to as Meta Compute, according to The Wall Street Journal. The company has not announced a public cloud service under that name.
The company is also hiring Dave Brown, a senior Amazon Web Services executive who spent nearly two decades at Amazon. Brown is expected to report to infrastructure head Santosh Janardhan and work on Meta’s data centre expansion.
Compute lease or cloud platform?
The reported negotiations concern access to computing capacity rather than the launch of a complete public cloud platform. The available reporting does not state whether Meta would provide dedicated clusters, customer-facing management tools, operational support, or broader managed infrastructure services.
Established cloud providers combine processor capacity with storage, databases, networking, identity controls, security tools, billing, service-level agreements, and technical support. Meta has not announced a comparable external service portfolio.
Separate reports have said Meta is considering commercial access to both computing capacity and AI models. Those would be different offerings, and there is no indication that Meta’s own models would form part of the proposed Anthropic agreement.
A completed deal would place Meta on both sides of the infrastructure market, buying capacity from specialist providers while leasing some of its own resources to another AI company. It would also give Meta experience supplying large-scale computing infrastructure to an external customer without establishing the company as a full hyperscale cloud provider.
Reuters reported that a commercial compute business would place Meta in competition with specialist infrastructure providers such as CoreWeave and Nebius. AWS, Microsoft Azure, and Google Cloud offer wider portfolios that include databases, storage, developer tools, security products, and managed AI services.
The reported agreement provides too little information to assess any effect on enterprise AI infrastructure prices. Neither company has disclosed the number or type of processors involved, expected utilisation, service terms, or pricing per unit of capacity.
It is also unclear whether Meta intends to negotiate private agreements with a limited number of large AI developers or make computing capacity available to a broader enterprise customer base. Private infrastructure contracts would not necessarily provide the price transparency or on-demand access associated with public cloud services.
Without published service terms, broader customer availability, or unit pricing, the proposed arrangement cannot be directly compared with enterprise offerings from hyperscale cloud providers or specialist infrastructure operators. The companies have not disclosed when the negotiations are expected to conclude.
(Photo by Muhammad Asyfaul)
See also: Data centre delays expose AI cloud power limits
Want to learn more about Cloud Computing from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
Cloud Computing News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars.