Meta Platforms is in preliminary discussions with artificial intelligence startup Anthropic regarding a substantial computing power lease arrangement that could be worth as much as $10 billion over a two-year timeframe, according to reporting by the New York Times citing three individuals familiar with the negotiations. The discussions underscore Meta's emerging ambitions to monetise its vast technological infrastructure beyond its core advertising business, a strategic pivot that Chief Executive Mark Zuckerberg has previously signalled as a priority for the company's future revenue streams.
The potential agreement would see Anthropic, the AI research company behind the Claude language model, lease computational resources from Meta at monthly intervals, with payments distributed across the two-year period. However, both parties have maintained flexibility in the arrangement, with either company able to terminate the agreement early should circumstances warrant such action. The structure reflects the still-evolving nature of cloud computing partnerships within the artificial intelligence sector, where pricing models and service arrangements remain fluid as the market matures.
For Meta, this deal represents a critical step in diversifying its income sources at a time when global advertising markets face headwinds and investor scrutiny over the company's ability to grow beyond its traditional model. By converting its significant capital investments in artificial intelligence infrastructure into a revenue-generating service, Meta would compete directly with specialised firms such as CoreWeave and Nebius that have emerged to serve the explosive demand for computational capacity driven by the advancement of large language models and other sophisticated AI tools. This strategic reorientation acknowledges that computational power itself has become a scarce and valuable commodity in the artificial intelligence economy.
According to the New York Times report, Anthropic initiated the proposal in June, approaching Meta with the lease opportunity as the startup prepares for an anticipated initial public offering. The negotiations, however, have encountered obstacles stemming from Meta's lack of established infrastructure for commercialising its computing resources, as the company has not previously operated as a vendor of computational services. This operational gap suggests that should the deal proceed, Meta would need to establish new business divisions, billing systems, and service level agreements specifically designed to serve external customers rather than its own product needs.
The timing of these discussions aligns with broader industry trends regarding the concentration of computing resources. In May, Anthropic secured a separate agreement with SpaceX to utilise the full computational capacity of Colossus 1, the data center facility located in Memphis, Tennessee. That arrangement demonstrated the intense competition among major artificial intelligence developers to secure access to processing power sufficient to train and operate increasingly sophisticated models. Meta's potential entry into this competitive space would fundamentally alter the dynamics of the market for computational resources.
Market reaction to the reporting was subdued, with Meta's shares declining more than two percent during trading on the day the story emerged, though the company's losses were somewhat arrested by the positive sentiment around the infrastructure diversification strategy. This modest market response reflects investor uncertainty about whether Meta can successfully execute a cloud computing business despite possessing the underlying technological assets and scale. For Malaysian and Southeast Asian markets, such arrangements remain largely abstract concepts, though they ultimately influence the availability and pricing of AI services accessible to regional businesses and startups.
During Meta's shareholder meeting held in May, Zuckerberg had explicitly acknowledged the company's interest in pursuing cloud computing opportunities, noting that Meta receives regular inquiries from other organisations seeking to purchase either access to its proprietary artificial intelligence models or surplus computational capacity. This public statement indicated that Meta had already begun receiving unsolicited proposals from prospective customers, providing preliminary validation of market demand for the types of services the company might offer. Zuckerberg's remarks suggested that entering this market was not merely speculative but responsive to genuine commercial interest from other technology firms.
Separately, Bloomberg News had reported earlier in July that Meta was actively constructing the infrastructure and organisational framework for a dedicated cloud business focused specifically on selling excess computational power and offering hosting services for artificial intelligence models developed by external software developers. This reporting suggests that the Anthropic discussions, rather than representing an isolated negotiation, form part of a broader strategic initiative to systematically transform Meta's infrastructure into a revenue-generating platform. The company's vast computational investments, originally made to power its own recommendation algorithms and content moderation systems, could potentially serve dozens or hundreds of external customers seeking access to cutting-edge artificial intelligence capabilities.
Both Meta and Anthropic have declined to provide detailed comment beyond confirming that discussions are occurring. Meta did not respond immediately to inquiries from Reuters, while Anthropic similarly offered no substantive remarks, maintaining the confidentiality customary during early-stage commercial negotiations. Reuters was unable to independently verify the specific financial figures and timeline cited in the New York Times reporting, highlighting how information about major technology negotiations frequently circulates through selective media channels before formal announcement.
The preliminary nature of these discussions merits emphasis, as the New York Times reporting explicitly noted that negotiations remain in their earliest phases and may ultimately fail to produce a binding agreement. Various factors could derail the arrangement, including disagreements over pricing, service level commitments, intellectual property considerations, or shifts in either company's strategic priorities. For observers monitoring the competitive landscape of artificial intelligence infrastructure, the significance of these discussions lies not in any particular deal outcome but rather in confirming that major technology platforms increasingly view their computational assets as distinct business opportunities rather than merely as internal support systems.
Looking ahead, the successful completion of such a deal would represent a watershed moment for the cloud computing market within the artificial intelligence sector, validating the notion that hyperscale technology companies can profitably serve as infrastructure providers for other ambitious artificial intelligence ventures. Whether Meta and Anthropic reach agreement remains uncertain, but the very fact of their engagement signals that the structural economics of artificial intelligence development increasingly depend on specialised computational markets where access to processing power becomes a distinct commodity with negotiated terms, pricing structures, and service arrangements.
