Meta discusses leasing up to $10B in AI computing capacity to Anthropic, report says

Meta Platforms is reportedly in discussions with Anthropic about a large-scale arrangement to lease computing capacity valued at as much as $10 billion. The proposal, as described by Yahoo Finance in a report syndicated through CryptoProwl, would involve Meta supplying cloud-like AI infrastructure on a capacity basis rather than selling a one-off hardware shipment.

The figure, up to $10 billion, indicates the magnitude of demand Meta and other infrastructure providers are facing as frontier AI models drive a race for GPUs, power, and data-center space. For Anthropic, outsourcing capacity in the near term can reduce the need to build all of the compute stack internally at the same pace as model development, while preserving flexibility as demand shifts by use case and model size.

For Meta, an agreement of this scale would represent a way to monetize its infrastructure buildout beyond its own AI systems and internal training and inference workloads. Leasing capacity to external partners also spreads utilization risk. If demand from Meta’s own AI projects fluctuates, external leasing can smooth revenue, at least in theory, while also strengthening relationships with AI developers that may become future customers for additional capacity.

The report does not specify which parts of the compute stack would be included in the lease, such as training versus inference capacity, the geographic footprint of data centers, or whether the deal would be priced on contracted capacity, consumed usage, or a hybrid model. It also does not disclose the expected timeline for negotiations or any definitive sign-off from either company.

Meta did not make a corresponding public announcement in the materials surfaced here. Its Newsroom provides a stream of official company updates, but the existence and terms of any $10 billion AI infrastructure leasing plan were not confirmed by Meta in the information available for this review.

The broader sector context is that AI infrastructure is increasingly sold and contracted as a service. Companies that control data centers and high-performance hardware can package capacity into offerings for model developers and enterprise AI customers, turning compute availability into a competitive advantage. Large language model providers, meanwhile, often need rapid access to large clusters, and partner arrangements can shorten time to scale.

Still, major uncertainties remain. The report does not confirm whether the parties have reached a binding agreement, the specific duration of the proposed capacity lease, or the technical configuration and service-level commitments that would govern uptime, performance, and capacity guarantees.

What to watch next is whether either company issues a formal statement, whether details emerge in a regulatory filing or investor communication, and how the market interprets the economics of leasing AI capacity at multi-billion-dollar scale. If negotiations progress, industry analysts will likely focus on contract structure, duration, and whether the arrangement reflects a broader shift toward compute partnerships across the AI ecosystem.