Meta's Pivot to Cloud Leasing Sparks Concerns of Computing Power Oversupply. Micron Falls Nearly 10%, Marvell Plunges 7%: Is the Logic Behind AI Hardware Stocks Shaken?

TradingKey - On July 1, Eastern Time, AI hardware stocks came under pressure, with memory and optical communication stocks plummeting across the board. SanDisk ( SNDK) fell 10.82%, Micron Technology ( MU) fell 9.7%; Corning ( GLW) dropped over 13%, Marvell Technology ( MRVL) declined over 7%, and Lumentum ( LITE) fell over 6%. It is reported that Meta plans to enter the cloud computing market, transforming from a pure computing power buyer into a computing power lessor.

According to media reports, Meta is officially planning its AI cloud infrastructure business, transitioning from a pure computing power buyer to a market participant with supply capabilities. It is simultaneously laying out two business lines: model services and bare-metal computing power leasing, directly targeting the three traditional cloud giants—AWS, Azure, and Google Cloud—while posing a disruptive threat to vertical AI computing power providers such as CoreWeave.

Under the plan, Meta's cloud business will adopt a dual-track model: first, a Model-as-a-Service route, referencing the platform logic of AWS Bedrock, which will host self-developed large models like Muse Spark based on its own AI infrastructure, opening APIs to developers and billing them based on usage; second, the leasing of raw computing power, following the path of emerging cloud providers to export bare-metal computing resources from its data centers and chips.

Currently, the demand for AI computing power continues to explode, and the commercial viability of computing power leasing has been proven. Relying on SpaceX's infrastructure resources, xAI quickly entered the market and secured major clients such as Anthropic, with its long-term revenue projection of $100 billion validating the feasibility of tech giants exporting computing power.

However, Meta's official announcement of selling its excess AI computing resources may overturn the logic of tight AI computing power demand. Previously, the market generally believed that against the backdrop of strong demand for computing power, upstream AI-related hardware would be frantically snapped up. Yet, Meta's decision to lease out its surplus computing power suggests that supply has exceeded demand. This might imply that Meta's new procurement volume of memory chips, HBM, and other hardware will shrink significantly.

However, there are also highly divided views in the market, with some arguing that the decline in the memory and optical communication sectors this time stems from a misinterpretation by the market.

Analysts state that although Meta has stockpiled large-scale computing resources, it lacks industry-competitive AI large model products of its own. Consequently, its internal business demand is insufficient to digest all of its existing computing power, leading to the decision to lease the surplus computing power to third-party leading AI companies.

If this commercial model is successfully validated and executed, Meta will not only refrain from scaling back its AI hardware procurement, but is also expected to accelerate its data center layout to grab cloud service market share, further driving new demand for upstream hardware such as storage and chips.