Recent developments from Meta and Anthropic have triggered a re-evaluation of the AI investment landscape, signaling a shift from a focus on expanding capital expenditure to prioritizing capital efficiency. Meta is reportedly exploring ways to commercialize its surplus AI computing power by leasing it to external customers or offering AI models deployed on its infrastructure. Simultaneously, Anthropic is in discussions with Samsung Electronics to develop its own AI chips, potentially using a 2-nanometer process to reduce long-term computing costs and lessen reliance on single suppliers. These actions combined suggest that AI companies are now seeking to improve the return on investment of existing infrastructure rather than simply continuously expanding it, leading the market to question the pricing structure of the entire AI trade.
The market reaction has been swift and severe, particularly for the AI hardware sector. Memory stocks and AI semiconductor stocks have experienced significant declines. Goldman Sachs' basket of memory stocks fell over 18% in two days, marking its sharpest two-day decline in 12 years. Similarly, AI semiconductor stocks suffered their worst two-day move since "Liberation Day." Overall, the Philadelphia Semiconductor Index (SOX) dropped over 10% cumulatively in two days, its worst two-day performance in nearly a month. This correction was not limited to a few overheated names but showed characteristics of a crowded trade being forced to shrink, with leaders falling sharply and short positions rallying, leading to a jump in realized volatility. Companies like SanDisk, Kioxia, Micron, SK Hynix, and Samsung all saw significant weakening, with some entering bear market territory.
Despite the sharp market correction, many institutions do not view these events as a denial of the AI super-cycle or a peak in AI demand. Instead, they interpret it as the AI industry transitioning from a "battle of capital expenditure" to a "battle of capital efficiency." This change in expectations served as a catalyst for the AI hardware sector correction. Analysts believe that Meta's move to sell surplus computing power is aimed at finding a commercial outlet for its massive AI capital expenditure to enhance the sustainability of future investments, rather than cutting spending. The long-term AI story remains intact, with continued demand for AI infrastructure, but the burden of proof for justifying high valuations has increased. The market is now looking for companies that can deliver orders, margins, and spending momentum to validate their valuations.
While the market saw a rotation of funds away from technology and utilities, with financials and consumer discretionary outperforming, the overall S&P 500 remained near its highs. Hyperscalers are still spending heavily, with AI capital expenditure from major players like Meta, Microsoft, Amazon, and Alphabet projected to reach $725 billion this year. Meta, for instance, raised its full-year capital expenditure range to $125 billion to $145 billion, with its CFO noting that the company has "continued to underestimate our compute needs." This indicates that the buildout of AI infrastructure is far from mature, and the demand for computing power and token consumption for large models continues to exceed expectations. However, the market is no longer willing to pay for every part of the AI story in the same way, suggesting a more discerning approach to investment in the sector.
