Goldman Sachs Trading Desk: Momentum Trading May Take Weeks to Bottom Out, AI Capital Expenditure Narrative Is Shaking

AI models have achieved cutting-edge performance with far less computational power than the market expected, once again challenging the prevailing narrative that “only by continually expanding capital expenditures can one win the AI competition.” According to the Goldman Sachs One-Delta Trading Desk, momentum strategy adjustments are not yet over, but currently, there are no signs of systemic risk in the US stock market, and market structure remains resilient.

Rich Privorotsky, head of the Goldman Sachs One-Delta Trading Desk, noted that their momentum model indicates the current momentum trading (Momentum) pullback likely still has several weeks to go before bottoming out, and the eventual drawdown is likely to be close to the historical median level. However, considering the previous rate of increase was much steeper than the historical average, there remains a possibility that this correction will exceed historical averages.

He also pointed out that the emergence of a new generation of highly efficient AI models is prompting the market to rethink the investment logic behind AI infrastructure. As model training efficiency continues to improve, the narrative of “continuously pouring huge capital into building ever-larger computing clusters” is facing increasing skepticism, while AI capital expenditures remain the main pricing driver for global tech stocks.

Momentum Trades Not Yet Fully Cleared, Internal Market Rotation Persists

Privorotsky stated that his tracked momentum indicators show that relative volatility remains high, and so far no signal has appeared that would alleviate concern.

However, clear divergences have started to emerge within the market. On the one hand, some AI hardware-related stocks have entered oversold territory; on the other, some previously underperforming sectors have rebounded even without significant fundamental improvements, with evident signs of capital rotation.

From an index perspective, US equities as a whole continue to show strong resilience. Correlation among sectors remains low, with capital largely shifting between industries rather than triggering broad-based sell-offs. Even though implied volatility rose somewhat last Friday, the underlying market structure has not fundamentally changed.

AI Efficiency Gains Again Challenge CapEx Logic

Privorotsky said he was impressed by the engineering capabilities of the Kimi K3 model after actual testing. This model features 2.8 trillion parameters and self-hosting it still requires enterprise-grade GPU clusters—it is not something that can run on ordinary local devices.

He believes the real focus should not be on the inference phase, but rather on training efficiency.

Compared to models that solely depend on larger computing power, the new generation models increasingly rely on algorithmic optimization, architectural innovation, and more efficient Mixture of Experts (MoE) routing mechanisms to enhance training efficiency. For example, Kimi K3 possesses 896 expert modules, yet each inference only activates 16, substantially reducing computational resource consumption.

This has also prompted the market to reconsider: If frontier models can significantly improve training efficiency through algorithmic innovation, does the AI industry still need to continually build ever-expanding, capital-intensive data centers and training clusters?

However, Privorotsky believes this mainly challenges the investment logic on the training side, while demand for inference-related computing power remains strongly supported and the long-term demand for AI infrastructure has not seen a fundamental reversal.

Earnings Season Will Decide Whether the AI Narrative Continues

As the Federal Reserve enters its pre-meeting blackout period, markets will turn their short-term attention to macro events such as the ECB meeting, UK CPI, and preliminary PMI data from major global economies.

Nevertheless, Privorotsky believes that the true determinant of market direction will be the upcoming earnings season.

Besides Alphabet, results from Tesla, Texas Instruments, Intel, and AMD's upcoming “Advancing AI” event will serve as important windows for the market to observe the AI investment cycle and to further test whether the trillion-dollar logic of AI capital expenditures can continue to gain market acceptance.