Momentum Trade Bottoming Out May Take Several More Weeks! Goldman Gives U.S. Stocks a Confidence Boost: No Systemic Risks in Sight

An artificial intelligence (AI) model trained with significantly lower computational power than market expectations has once again shaken Wall Street's long-held narrative that 'only continuous capital expenditure can win the AI race.' However, Goldman Sachs (GS-US) recently assessed that although the momentum trade correction in tech stocks is not yet over, the broader U.S. stock market has not shown signs of systemic risk, and market structure remains resilient.

Rich Privorotsky, head of Goldman Sachs' One-Delta trading desk, noted that internal momentum models suggest the current pullback in momentum trading may still require several more weeks to reach a true bottom, with the final decline likely approaching historical median levels.

However, he cautioned that given the current rally's slope far exceeds past averages, the correction could potentially exceed historical norms.

Privorotsky observed that current momentum indicators still show relatively high volatility, with no clear signal yet for the market to ease its guard.

Nonetheless, clear divergence is already emerging within the market. Some AI hardware-related stocks have fallen into oversold territory, while lagging sectors without fundamental improvement are rebounding, indicating visible fund rotation.

From a broad index perspective, U.S. equities remain resilient, with low sector correlation, suggesting capital is rotating between industries rather than escalating into a broad sell-off. Even though implied volatility briefly spiked last Friday (17th), this market structure has not fundamentally changed.

What is truly prompting market reevaluation is the emergence of a new generation of highly efficient AI models. Privorotsky revealed that after personally testing the Kimi K3 model with 2.8 trillion parameters, he was deeply impressed by its engineering design—though the model still requires enterprise-grade GPU clusters for deployment and is not feasible for general devices.

He emphasized that the real focus should not be on inference, but on the leap in training efficiency.

Rather than simply stacking massive computational power, next-generation models rely more on algorithmic optimization, architectural innovation, and more efficient Mixture-of-Experts (MoE) routing mechanisms.

For example, Kimi K3 embeds 896 expert modules but activates only 16 during each inference, drastically reducing computational resource consumption.

This advancement forces the market to reconsider a core question: if cutting-edge models can significantly improve training efficiency through algorithmic innovation, does the AI industry still need to continuously expand large-scale, capital-intensive data centers and training clusters?

Privorotsky believes this skepticism primarily impacts the investment logic on the training side, while computational demand on the inference side remains strong. The long-term need for AI infrastructure has not fundamentally reversed.

With the Federal Reserve entering its pre-meeting blackout period, short-term market focus will shift to the European Central Bank's rate decision, UK CPI data, and preliminary PMI readings from major global economies.

However, Privorotsky stressed that the key driver of market direction remains the upcoming earnings season.

Beyond Alphabet (GOOGL-US), earnings reports from tech giants like Tesla (TSLA-US), Texas Instruments (TXN-US), and Intel (INTC-US), along with AMD's (AMD-US) upcoming 'Advancing AI' event, will serve as critical indicators for the trajectory of the AI investment cycle, further testing whether the trillion-dollar AI capital expenditure narrative can continue to gain market support.