Anthropic turns to Meta for $10B in computing power before IPO

Anthropic has proposed leasing up to $10 billion of computing power from Meta Platforms over two years as the AI developer prepares for a possible October IPO.

According to Reuters, which cited The New York Times, Meta is reviewing the proposal after Anthropic presented the terms in June. The planned agreement would require Anthropic to make monthly payments for access to Meta’s computing capacity.

Both companies could end the contract before the two-year period expires, the report added. Meta and Anthropic have not finalized the arrangement, leaving its value and duration subject to the outcome of their talks.

For Anthropic, the lease would provide access to the processing capacity needed to train and operate advanced artificial intelligence models. AI developers depend on large numbers of specialized chips and data centers, making reliable computing access a central part of their expansion plans.

Meta, in turn, could earn revenue from infrastructure built primarily for its own AI products and services. According to the report, the proposed lease would give the social media company another way to generate returns from its computing investments beyond its advertising business.

Meta could enter the AI infrastructure market

A completed agreement would place Meta in competition with CoreWeave and Nebius, two companies that supply computing infrastructure for AI workloads. Reuters reported that the Anthropic proposal could turn Meta into a provider of capacity to an external AI developer while it continues building models and products of its own.

The talks have emerged as technology companies compete for chips, electricity and data center space. Under the reported structure, Anthropic would secure capacity from a company that has spent heavily on AI infrastructure, while Meta would add a potential customer for resources within its computing network.

Anthropic has also pursued separate long-term infrastructure arrangements. Earlier this month, the company signed a 20-year data center lease with Bitcoin miner TeraWulf. The agreement is expected to supply additional computing resources for Anthropic’s future AI development.

Taken together, the Meta discussions and TeraWulf lease show how Anthropic is assembling the infrastructure required to support its models. Any assessment of the scale or financial effect of those agreements, however, depends on their final terms and the amount of capacity Anthropic ultimately uses.

Anthropic could reach public markets in October

Bloomberg reported that Anthropic is moving forward with preparations for a possible stock market listing. Banks working on the offering have begun arranging meetings between company executives and prospective investors, according to the report.

Those meetings could support an IPO as early as October, although Bloomberg’s timeline remains subject to market conditions and the company’s final decision. An October debut would put Anthropic in the public market before OpenAI, which Bloomberg reported is considering a listing in 2027.

Chinese AI developer DeepSeek is also preparing for an eventual public offering, according to the original report. Anthropic could therefore become one of the first major companies from the latest generation of AI model developers to list its shares.

Before the IPO report emerged, Anthropic had received approval from the US government to restore access to its Mythos 5 model for selected companies and federal agencies last month. Combined with its infrastructure agreements and investor meetings, the decision adds another development for banks and potential shareholders to examine as preparations continue.