Meta Platforms is currently engaged in early-stage discussions to lease computing power to Anthropic, a potential agreement valued at up to $10 billion over a two-year term. According to reports from July 17, 2026, the proposal, initiated by Anthropic in June, outlines a structure involving monthly installments with provisions for early exit rights for both parties. While these talks remain preliminary and may not result in a finalized contract, the development signals a notable evolution in how hyperscalers manage their massive infrastructure investments.
The financial scale of this potential deal is significant, yet it sits within a broader, highly competitive landscape of compute procurement. Anthropic already maintains substantial commitments with other major players: a three-year agreement with SpaceX/xAI valued at approximately $45 billion, an Amazon partnership exceeding $25 billion with 5GW of capacity, and a Google/Broadcom arrangement reaching up to $40 billion with 3.5-4.5GW of TPU capacity. For Meta, which raised its 2026 capital expenditure forecast to a range of $125-145 billion, this deal would represent a departure from its current operational model. Meta does not currently sell compute to third parties; entering this space would establish an entirely new business line, effectively positioning the company to compete directly with neocloud firms like CoreWeave and Lambda.
Market participants reacted swiftly to the news. On July 17, Meta shares fell as much as 6% during intraday trading before paring those losses. Retail sentiment, which had been bullish, shifted to neutral within 24 hours of the report. This volatility reflects ongoing investor anxiety regarding the justification for the industry’s aggressive AI capital expenditure. Wall Street remains focused on whether these massive outlays will translate into sustainable revenue streams, making the prospect of a new, high-value lease arrangement a focal point for analysts evaluating Meta’s long-term capital efficiency.
The broader context for this demand is underscored by recent data from the semiconductor supply chain. TSMC’s Q2 2026 earnings, released on July 16, confirmed an acceleration in compute demand, with the firm raising its own capital expenditure guidance to $60-64 billion and projecting revenue growth above 40% year-over-year. As the hardware bottleneck persists, the pressure on hyperscalers to optimize their utilization rates intensifies.
The structural significance of these talks lies in the transition from a build-to-use model toward a build-to-scale-and-monetize framework. Historically, hyperscalers constructed massive data center footprints primarily to support their own internal product ecosystems. However, as the cost of maintaining this infrastructure climbs, the incentive to transform idle or excess capacity into a revenue-generating asset becomes increasingly compelling. If Meta successfully pivots to become a compute provider for third-party AI labs like Anthropic—which is reportedly planning an IPO for October 2026—it would mark a fundamental change in the business model of the world’s largest technology companies. This shift suggests that the future of hyperscale infrastructure will be defined less by proprietary isolation and more by the commoditization of compute as a primary service, effectively turning the massive capital investments of the last few years into the foundation for a new, diversified revenue layer.