AI capex is now a credit story: $269B of issuance meets $300B of autocallables

Charlie McElligott, Nomura's cross-asset macro strategist, published a note dated August 14, 2026 flagging more than $300 billion of autocallable structures sitting in the derivatives market, mostly linked to single-name mega-cap tech ^[1][11]. It lands alongside $269 billion of year-to-date debt issuance from AI companies, hyperscalers and datacenter operators, roughly 12 times the 2015-2024 annual average ^[2]. The consequence for operators is that the capex cycle is no longer only an earnings-quality question; it now runs through funding markets and volatility.

Start with the arithmetic. A 12x multiple on $269 billion implies the prior decade averaged about $22 billion a year from this cohort ^[15]. Broader corporate issuance is up 61% year over year ^[3], so the AI complex is not merely participating in a hot primary market, it is redefining the scale of it. At mid-August, $269 billion annualises to roughly $430 billion for the full year ^[16].

The comparison the source draws is worth handling carefully. Morgan Stanley had projected $250 billion to $300 billion of hyperscaler issuance for 2026 ^[4], and the publisher treats the $269 billion figure as having essentially met that range with more than four months to run ^[5]. That is 108% of the low end and about 90% of the high end ^[17], but the $269 billion tally covers AI companies, hyperscalers and datacenter operators together ^[2] while the forecast was for hyperscalers alone ^[4]. The direction is right; the precision is not.

McElligott's stated concern is not debt service ^[6]. It is the plumbing. Autocallables are exotic products sold to institutions and wealth channels that pay enhanced yields in exchange for exposure to an underlying stock, and they redeem automatically when the stock hits predetermined levels, forcing dealers to unwind hedges ^[7]. In calm markets the flow suppresses volatility: dealers hedge by selling options, which adds supply of volatility and pushes implied vol lower, which encourages more issuance ^[8]. McElligott calls the resulting positioning a "coiled spring" ^[10].

The unusual part is the trigger direction. If key tech names gap higher, call barriers breach at once, dealers buy back the options they had sold, and implied volatility can spike non-linearly, cascading through risk models and forcing selling elsewhere ^[9]. A melt-up, not a crash, is the initiating event. That also explains why the usual dashboard is unhelpful: the VIX prices aggregate 30-day S&P 500 expectations and does not reflect concentrated single-name exposure or the mechanics of a structured product unwind ^[12]. All of this is happening into a dovish backdrop of accommodative central banks, tight spreads and healthy risk appetite ^[11], which is precisely the regime that makes the position build.

Two things to watch. First, proximity of the large tech names to widely held barrier levels; McElligott's framing is that as Nvidia, Microsoft or Meta approach popular strikes, the probability of a cascading unwind rises materially ^[13]. Second, credit conditions, because the issuance machine keeps running only while spreads stay tight and investment-grade demand holds ^[14]. Companies planning to fund datacenter capex on 2026 terms should treat that second condition as the live variable, not the first.