Meta opens surplus computing power for commercial use in public cloud, AI capital expenditure logic changes, US stock hardware sector collectively under pressure

On July 2nd local time, Meta officially launched its surplus computing power external commercialization plan, opening up vast idle AI computing power from its self-built data centers for global enterprise leasing. This marks a key turning point in the years-long aggressive infrastructure investment cycle among North American tech giants, fundamentally reshaping the existing landscape of the global public cloud and semiconductor industry chains.

In recent years, Meta has made significant investments in data centers and GPU procurement to support its proprietary large models, with annual capital expenditures peaking at nearly $145 billion. This has resulted in massive clusters of computing power, with large amounts of redundant capacity beyond its own business needs. With the launch of its commercial computing power business, Meta is simultaneously opening up both bare computing power leasing and proprietary large model service invocation, directly entering an enterprise cloud track long dominated by Amazon AWS and Microsoft Azure. At the same time, it faces direct competition from specialized computing power rental providers such as CoreWeave. Leveraging its ultra-large-scale computing resources for natural cost advantages, Meta is set to quickly capture enterprise orders in the AI inference power market.

Capital markets showed an extreme divergence: Meta's share price soared 8.81% on the day, as investors saw the monetization of computing power as a new path to recouping high infrastructure spending, benefiting long-term cash flow. However, this quickly raised core concerns in the market: leading tech firms can now absorb their own computing power internally and generate external revenue, which will inevitably slow the pace of new GPU and memory chip procurement in the future. The valuation logic of long-term AI hardware shortages is being challenged. Driven by these pessimistic expectations, US semiconductor and memory sectors suffered heavy losses: Micron Technology plunged 10.57%, SanDisk fell 10.62%, TSMC's share price dropped nearly 7%, and the Nasdaq chip index weakened in tandem. Upstream chip and memory manufacturers' profit expectations significantly cooled.

Wall Street institutions are notably divided in their views. Optimists see this as a reasonable transformation to refined operations for tech companies, where revitalizing idle assets will optimize input-output ratios; while firms like Goldman Sachs warn that a surge in computing power supply flooding the market will continue to compress the bargaining power of hardware manufacturers, and that going forward global AI industry chain capital expenditure will shift from rapid expansion to cautious control. Looking ahead, Meta’s foray into cloud computing is set to propel the global computing power market from a “hardware buying race” to “computing power operations competition,” bringing a new round of reshaping to profit distribution models across the industry chain.