Nomura: AI capex caught in a dilemma—accelerating pressures cash flow, slowing means the growth story breaks

The AI investment narrative is facing an internal contradiction in its own logic. The market's judgment on the capex path of hyperscale cloud providers has evolved from the previous "heads I win no matter what" to the current "choose between two evils" dilemma—acceleration pressures free cash flow, while deceleration casts doubt on the demand outlook; both paths point to a selloff in AI-related assets.

Zhuifeng Trading Desk news, October 23, according to Nomura cross-asset strategist Charlie McElligott's report, over the past week, negative sentiment catalysts within the AI ecosystem emerged one after another, and combined with the long positions accumulated during the September "agent craze" rally, the market experienced large-scale forced liquidation.

In the most violent single-day selloff alone, of the 124 US equity long-short pairs monitored by Nomura, 47 saw single-day moves exceeding 3 standard deviations, and 6 hit extreme deviations of more than 5 standard deviations.

Next week the market will enter a dense disclosure window for hyperscaler earnings, including Google, Amazon, and Microsoft, combined with the Federal Reserve's interest rate meeting; market direction will face a key test in the short term.

The "prisoner's dilemma" of capex

McElligott described this situation in the report as "either choice is poison":

First, if hyperscaler capex continues to accelerate. As the trend revealed by Meta's latest earnings shows, the market will focus heavily on the continued depletion of free cash flow and further question overall financing sustainability.

Second, if hyperscaler capex shows a substantial deceleration, it means the "optimal scenario" already priced by the market—covering core theses such as AI demand, technology adoption, profitability, and end-user productivity gains—is breaking down, and the AI trade will face a more violent shock.

The formation of this dilemma is highly related to the rapid rebuilding of long positions during the September "agent concept" rally. At that time, large amounts of capital poured into core AI names, accompanied by sizable speculative call option positioning, laying the groundwork for the subsequent stampede-like exit.

Stop-loss triggers spark cross-market deleveraging

Crowded positioning is the core reason for the unusually violent volatility this round.

According to Nomura trading desk data, in the most violent single-day selloff, a total of 124 US equity long-short pairs triggered stop-losses, of which 47 pairs moved more than plus or minus 3 standard deviations, 26 pairs exceeded plus or minus 4 standard deviations, and 6 pairs exceeded plus or minus 5 standard deviations.

(Performance of various pair trades over the past period)

McElligott believes the root cause of this stop-loss wave is that throughout 2026, "high Sharpe ratio trades" were highly concentrated, and carry strategies and trend-following strategies resonated for a long time, forming a rare broad-based winner effect.

And the current point coincides with the end of the fourth quarter, when fund managers face year-end performance pressure. In an already stressed drawdown environment, the motivation to actively seek a rebound is far weaker than the defensive instinct to "preserve capital and cross the year safely"—a "de-risk, close positions" mode dominates market behavior.

According to Nomura QIS data, over the past two weeks, CTA trend strategies have closed about $68 billion of US Treasury futures shorts and $38 billion of short-term interest rate futures shorts, but the signal remains in a net short state, meaning that if the bond rally continues, the potential forced covering scale remains considerable.

(Over the past two weeks CTAs have completed more than $100 billion of short covering; to trigger the next wave of about $50 billion of strong buying in the green box, bond prices need to wait for a further upward breakout through the +2% trigger threshold)

McElligott believes that as the short-term window covering is basically complete, the squeeze momentum in bond shorts is pausing for now. Both the 2-week and 1-month cycle models have flipped to "long" signals, but for the 3-month cycle model to flip long from the current level, bond prices still need a substantial further rise.

Macro resonance: correlation switch between bonds and equities

On the bond side, the "negative stock-bond correlation" protective property has reappeared. McElligott pointed out that a large amount of capital completed an asset allocation switch on the previous trading day through selling equity futures and buying Treasury futures, and bonds once again played the role of a risk hedge.

This feature, after persisting for most of the year, had once failed, and now has returned as the credibility of the "growth scare" scenario rises.

The equity market showed the opposite switch. For most of 2026, the Nasdaq and semiconductor sectors exhibited a rare feature of "higher volatility on up days than on down days"; now, this relationship has reversed back to "negative spot-volatility correlation," meaning volatility is greater when stock prices fall.

This switch is most prominent in Nasdaq and SMH (semiconductor ETF) space, directly reflecting the market's repricing of AI capex sustainability.

(Volume on "down days" for the Nasdaq and semiconductor ETF indices is nearly 2:1 versus volume on "up days"; this phenomenon reappeared in October)

It is worth noting that VIX volatility at the overall S&P 500 index level is relatively mild. McElligott attributes this to the "reverse dispersion" effect—tech longs were hit hard, while rate-sensitive sectors and other short targets rebounded sharply; the two offset each other so that overall correlation at the index level did not rise explosively.

According to McElligott's observations, the following notable call buying activity occurred over the past week:

CoreWeave paid total premiums of about $69 million, involving notional option exposure of about $170 million;

Nebius Group paid total premiums of about $25.7 million, involving notional exposure of about $80 million;

Micron paid total premiums of about $20 million, involving notional exposure of about $85 million.

The above trades involved out-of-the-money call options expiring from March to April 2025, showing that some capital still maintains directional bets on a medium-term rebound in the AI sector amid the current turmoil.

Regarding potential stabilization conditions, McElligott believes the key is whether sellers of the volatility risk premium (VRP) are willing to follow the historical pattern of "shorting into rising volatility," especially 0DTE The stabilizing role of Gamma selling strategies in buying dips and selling rallies will be an important observational indicator for judging whether volatility can converge again.

For more detailed interpretations, including real-time commentary and frontline research, please join [Zhuifeng Trading Desk▪Annual Membership] Risk Disclosure and Disclaimer

Markets carry risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this is at the user's own risk.