AI models achieving cutting-edge performance with far less computing power than the market anticipated are once again challenging the prevailing narrative that "only continuous expansion of capital expenditure can win the AI race." The Goldman Sachs One-Delta trading desk believes the adjustment in momentum strategies is not yet over, but the current U.S. stock market is not facing systemic risk, and market structure remains resilient.
Rich Privorotsky, head of the Goldman Sachs One-Delta trading desk, stated that their momentum model indicates it may still take several weeks for the current momentum trade correction to truly bottom out, with the final decline likely approaching historical median levels. However, given the previous ascent was much steeper than the historical average, this adjustment also has the potential to exceed historical norms.
He also pointed out that the recent emergence of a new generation of highly efficient AI models is prompting the market to re-evaluate the investment logic for AI infrastructure. As model training efficiency continues to improve, the core narrative of "continuously investing massive capital to build ever-larger computing clusters" is facing increasing scrutiny, while AI capital expenditure remains the most critical pricing theme for global tech stocks.
Momentum Unwinding Not Yet Complete, Rotation Persists Within Market
Privorotsky noted that their tracked momentum indicators show relative volatility remains elevated, with no clear signals yet to sound the all-clear.
Nevertheless, significant divergence is emerging within the market. On one hand, some AI hardware-related individual stocks have entered oversold territory. On the other hand, some previously underperforming sectors are rebounding without a clear improvement in fundamentals, indicating a clear characteristic of capital rotation.
At the index level, the overall U.S. stock market continues to show strong resilience. Correlations between sectors remain low, with capital primarily shifting between industries rather than devolving into a broad-based sell-off. Even with the rise in implied volatility last Friday, this market structure has not fundamentally changed.
AI Efficiency Gains Question Capital Expenditure Rationale
Privorotsky stated that after practical testing of the Kimi K3 model, he was impressed by its engineering capabilities. The model has 2.8 trillion parameters and still requires an enterprise-grade GPU cluster for self-hosting, not something that can run on ordinary local devices.
He believes the truly noteworthy aspect is not the inference stage, but training efficiency.
Instead of relying solely on larger-scale computing power, the new generation of models focuses more on improving training efficiency through algorithm optimization, model architecture innovation, and more efficient Mixture-of-Experts (MoE) routing mechanisms. Taking Kimi K3 as an example, it has 896 expert modules but activates only 16 of them during each inference, significantly reducing computational resource consumption.
This is prompting the market to reconsider: if cutting-edge models can significantly enhance training efficiency through algorithmic innovation, does the AI industry still need to continuously build ever-expanding, capital-intensive data centers and training clusters?
However, Privorotsky believes this challenges the investment logic for the training side more, while the demand for computing power on the inference side remains robustly supported. The long-term demand for AI infrastructure has not fundamentally reversed as a result.
Earnings Season to Determine AI Theme's Sustainability
As the Federal Reserve enters its quiet period ahead of its policy meeting, the market's short-term focus will shift to macro events such as the European Central Bank's policy meeting, UK CPI data, and preliminary PMIs for major global economies.
However, Privorotsky believes the upcoming earnings season will be the true determinant of market direction.
Beyond Alphabet, the earnings reports from tech companies like Tesla, Texas Instruments, and Intel, along with AMD's upcoming "Advancing AI" event, will serve as crucial windows for the market to observe the AI investment cycle. They will further test whether the trillion-dollar AI capital expenditure narrative can continue to gain market acceptance.
Comments