AI Capital Expenditure Faces a Dilemma: Accelerating Puts Pressure on Cash Flow, Slowing Down Undermines the Growth Narrative

The AI investment narrative is facing a fundamental contradiction in its own logic. The market's assessment of the capital expenditure paths for hyperscale cloud providers has shifted from the previous stance of "positive outcomes always win" to the current dilemma of "choosing the lesser of two evils"—accelerating capital expenditure puts free cash flow under pressure, while slowing down casts doubt on future demand prospects, with both paths leading to sell-offs in AI-related assets.

According to reports from Wind Chasing Trading Desk, on October 23, Charlie McElligott, cross-asset strategist at Nomura Securities, noted that in the past week, a series of negative sentiment catalysts have emerged within the AI ecosystem. Coupled with the heavily built long positions during the "agent fever" in September, the market has experienced widespread forced unwindings.

During the most violent single-day sell-off, out of 124 U.S. equity long-short pairs monitored by Nomura, 47 saw daily moves exceeding 3 standard deviations, with 6 experiencing extreme deviations beyond 5 standard deviations.

This volatility shocked not only core AI concept stocks but, through chains of forced unwinding in crowded positions, spilled over into U.S. Treasuries, exchange rates, and commodity markets, triggering cross-asset systemic deleveraging.

Next week, the market will enter a dense window of major quarterly earnings releases from hyperscale players, including Google, Amazon, and Microsoft, alongside the Federal Reserve's policy meeting. The market's direction will face a critical test in the short term.

The core narrative of AI trading has rapidly shifted from the previous one-sided logic of "positive outcomes benefit all" to the current two-way dilemma of "there's negative impact no matter what."

McElligott describes this predicament in his report as "either choice is poisonous":

First, if hyperscale companies continue to accelerate capital expenditure. As revealed in Meta’s latest earnings report, the market will intensely focus on the ongoing drain of free cash flow, further questioning the overall sustainability of financing.

This path carries nonlinear risk, implying the need to issue even larger-scale, higher-spread debt. AI trades come under pressure due to an "unsustainable" narrative.

Second, if capital expenditure by these large-scale firms materially slows down, it signals that the “best-case scenario” previously priced in by the market—covering AI demand, technology adoption, profitability, and end-user productivity gains—is falling apart, leading to even fiercer shocks for AI trades.

This dilemma is closely connected to the rapid rebuilding of long positions during the September "agent concept" rally. At that time, large amounts of capital poured into core AI assets, accompanied by aggressive speculative build-up in call options, paving the way for subsequent stampede-style exits.

Crowded positioning is the core reason behind the exceptional volatility in this round.

According to Nomura’s trading desk data, during the most intense single-day sell-off, a total of 124 U.S. long-short equity pairs triggered stop-losses; of these, 47 experienced swings greater than ±3 standard deviations, 26 exceeded ±4 standard deviations, and 6 exceeded ±5 standard deviations.

(Recent performance across different pair trading strategies)

McElligott believes the root of this stop-loss wave lies in the highly concentrated "high Sharpe ratio trades" throughout all of 2026, with carry trades and trend-following strategies resonating over time, creating a rare broad-spectrum winner effect.

Moreover, as the end of Q4 approaches, fund managers face year-end performance pressures. In a drawdown environment, the motivation to actively chase rebounds pales in comparison to the defensive instinct to “preserve capital and safely cross the year”—thus, the "de-risk and close positions" mode dominates market behavior.

At the same time, the AI narrative reset is accelerating systematic trend strategy position adjustments, especially those in previously extremely crowded U.S. Treasury and short-term rate shorts.

According to Nomura QIS data, over the past two weeks, CTA trend strategies have closed around $68 billion in U.S. Treasury futures shorts and $38 billion in short-term rate futures shorts, yet signals remain net short—meaning if the bond market rally continues, the potential for further forced covering remains significant.

(Over the past two weeks, CTAs have covered over $100 billion in shorts; for the next wave of strong buying, roughly $50 billion in the green box, a further +2% price breakout is required to trigger it)

McElligott believes that with short-term covering largely complete, the squeeze momentum on bond shorts is pausing at this stage. Models for 2-week and 1-month periods have turned long, but for the 3-month cycle model to flip long from here, a significant further rise in bond prices is required.

This AI narrative reset has simultaneously triggered structural changes in internal correlations between the bond and equity markets, which is a noteworthy signal at the market mechanism level.

On the bond side, the "negative equity-bond correlation" protection feature is re-emerging. McElligott points out that a large volume of capital completed asset allocation transitions via stock index futures selling and Treasury futures buying the previous trading day, with bonds once again acting as a risk-hedging tool.

This trait, which persisted for most of the year before failing at one point, is now returning as the credibility of a "growth scare" scenario rises.

The stock market is showing the opposite switch. For most of 2026, the Nasdaq and semiconductor sectors showed the rare phenomenon of "higher volatility on up days than down days;" currently, that relationship has reverted to a "negative spot-volatility correlation," meaning price drops now see higher volatility.

This switch is most pronounced in the Nasdaq and SMH (semiconductor ETF) spaces, directly reflecting the market's repricing of AI capital expenditure sustainability.

(Volume on Nasdaq and semiconductor ETFs during "down days" is nearly 2:1 compared to "up days," a phenomenon reappearing in October)

Notably, VIX-level volatility for the overall S&P 500 index has remained relatively mild. McElligott attributes this to a "reverse dispersion" effect—technology longs suffered sharp hits, while rate-sensitive sectors and other shorts rebounded abruptly, so the netting effect kept overall index-level correlation from surging explosively.

Despite overall pressured sentiment, several noteworthy contrarian positioning signals have appeared in the options market, mainly reflected in call option buying on select core AI names.

According to McElligott’s observations, the following significant call option buying activities were seen over the past week:

• CoreWeave paid approximately $69 million in premiums, with notional option exposure of about $170 million;
• Nebius Group paid roughly $25.7 million in premiums, with notional exposure of about $80 million;
• Micron paid a total of $20 million in premiums, with notional exposure of around $85 million.

The above trades involve out-of-the-money call options expiring in March through April 2025, indicating that despite the current turmoil, some investors continue to bet directionally on a medium-term rebound in the AI sector.

For potential stabilization conditions, McElligott believes the key lies in whether volatility risk premium (VRP) sellers are willing to stick to historical patterns of "selling volatility as it rises." Particularly, the consistency of 0DTE Gamma selling strategies in buying the dip and selling the rally will be a key indicator to observe whether volatility will converge again.

(0DTE Gamma range)