Bank of America: Meta selling computing power aims to tell a better AI investment return story

Author | Zhang Yaqi

    Editor | Hard AI

Meta is planning to monetize its massive AI computing power assets, a layout that is not only the prototype of a new business line but also a strategic signal responding to investor concerns about returns on high capital expenditures.

According to reports, Meta is formulating plans to launch a cloud infrastructure business that would provide external customers with access to AI computing power and model services. Following the announcement, Meta’s shares surged by about 10% in a single day, far outpacing the S&P 500’s increase of around 0.25%, indicating positive market response to this potential new business line.

According to Wind Trading Desk, Bank of America Securities analysts Justin Post and Nitin Bansal stated in a July 1 research report that the development of the cloud business helps highlight the potential value of Meta’s computing assets and model R&D, thus easing investors’ concerns over continued AI infrastructure investments without visible returns. Bank of America maintains a ‘Buy’ rating for Meta with a target price of $835.

Meta Cloud Plan Surfaces: Advancing Along Two Paths

According to Bloomberg, citing sources, Meta’s cloud business plan currently follows two directions: First, to provide AI model hosting services, allowing developers access to various models running on Meta’s existing AI infrastructure, including the Muse Spark series, with fees based on access volume—this is similar to Amazon AWS’s Bedrock product; second, to sell raw computing capacity directly, with a positioning closer to emerging cloud computing providers such as CoreWeave.

This plan falls under Meta’s internal strategic initiative called “Meta Compute,” focused on the construction and management of AI infrastructure. Meta’s CEO has publicly hinted at commercial opportunities in the enterprise market and has stated that the company is poised to sell computing power externally at prices higher than the cost of construction.

Bank of America’s report points out that from a macro perspective, if Meta’s capital expenditures in 2026 can support up to 3GW computing construction (estimated at $4–4.5 billion per GW), the establishment of a cloud platform in the near term will give the company greater strategic flexibility—In the event of surplus capacity, it may be leased externally at a price of $1–1.5 billion per GW annually, providing positive support for the company.

Doubts on Competitive Positioning, Strategic Controversies Hard to Avoid

Despite the strong market response, Bank of America bluntly pointed out potential doubts. Meta’s self-developed chips appear to be lagging behind mature hyperscale cloud providers such as Amazon, Microsoft, and Google. Meanwhile, the company is still actively purchasing computing power from third parties—including a recent 1.6GW purchase agreement with Crusoe.

This has led the market to question Meta’s strategic rationale: Can a company that still needs to purchase external computing power convincingly build a computing resale business? How should its competitiveness in the hyperscale cloud market be positioned?

Bank of America believes,Whether Meta can gain stronger market recognition in this field depends to some extent on the advancement of its large language model (LLM) capabilities—the more advanced the models, the higher the external demand for Meta’s computing resources, and the more solid the business logic of its cloud operations.

Improvements in AI Unit Economics: a Double-Edged Sword for Cloud Providers

Beyond Meta’s cloud plan, there are also signals of note on the cost side of AI computing power. According to The Information, OpenAI has reportedly found a system-level optimization that reduces the inference cost of specific models by about half. This optimization is achieved by utilizing existing server infrastructure more efficiently without new hardware or new model architecture. Reportedly, OpenAI has applied this optimization to non-logged-in ChatGPT traffic, which requires only a few hundred Nvidia GPUs to support. The specific principle and whether it can be extended to logged-in users, API workloads, or compute-intensive inference products remains unclear.

Bank of America believes,Improvements in computational cost efficiency are overall positive for large internet companies:If this technology can be scaled across the industry, it will increase the effective output of existing computing capacity with no further hardware investment required, alleviate the urgency of new capex, and improve the unit economic model of AI business operations. As agentic application scenarios drive a sharp increase in token consumption, the strategic value of computing power optimization will become even more prominent.

However, for hyperscale cloud service providers, declining inference costs do pose some pricing pressure risks. Meanwhile, better gross margin structure and a broader addressable market are expected to drive continued growth in AI workload demand, making the overall impact still positive.