The AI Unwind And Warsh's Long-End Gamble | Weekly Roundup

The AI trade unwind accelerated sharply after Ash Branner's fund, which had grown from roughly 225 million dollars to billions through AI-focused investing, was forced to liquidate its entire public markets position along with some private holdings including Anthropic. Ken Griffin and Citadel acted as the single buyer of those liquidated positions, and markets saw a significant reversion higher once the forced seller was removed. Sharks had begun shorting AI-related names the prior week once rumors circulated that Branner's firm was shopping for buyers.

Several compounding dynamics drove the severity of the unwind. AUM in 3x leveraged semiconductor ETFs had grown from roughly 25 to 30 billion dollars up to 100 billion dollars, producing daily rebalancing swings of 4 to 7 percent with no normal 1 percent days. Korean margin call activity added further pressure, and SK Hynix reported the first meaningful earnings miss in the semiconductor complex during this period. The episode illustrated that price drives narrative rather than the reverse, as many competing explanations circulated before the leveraged liquidation cause became clear.

A V-shaped recovery back to all-time highs in semiconductor names is considered unlikely given the leverage required to reach those levels, with SanDisk needing to approximately double from current levels to return to its prior high. The more probable outcome following the mean reversion bounce is an extended period of low-volatility chop, consistent with how damaged buyer bases in aggressively uptrending sectors like metals and crypto have historically behaved. Leopold's thesis on AI may ultimately prove correct, but path dependency and trade implementation caused losses regardless of directional accuracy, and markets can vindicate a thesis over years while still producing near-term losses.

Kevin Warsh's second Fed meeting came in with pre-meeting odds of roughly 60 percent for a pause and 40 percent for a hike, and the Fed delivered a pause with three dissents. Long-end yields and stocks turned lower together around 3pm while Warsh was still speaking, and markets subsequently priced in nearly two rate increases over the next 12 months. Warsh's core signal was that the long end of the yield curve, not the front end, is where the primary distortion lies. The Fed doubled its balance sheet during COVID and never fully unwound it, and the resulting record flat yield curve reflects ongoing balance sheet accommodation suppressing long-end yields. Warsh explicitly asked the committee how much accommodation the balance sheet is providing the long end, and his intent was to remove the Fed's footprint from the Treasury market and allow the long end to price to fair market value.

Allowing that long-end repricing would likely add 50 to 75 or possibly 100 basis points to long-end yields and would produce a stock market correction, lower growth, and lower inflation without requiring a front-end rate hike. Warsh chose to target the long end rather than hike the front end because the front end, with inflation around 3 to 3.5 percent and rates at 350 to 375 basis points, is only slightly loose, while the back end is where the distortion is most severe. Credit spreads widened following the press conference, consistent with the balance sheet tightening signal. Warsh was guiding markets toward balance sheet reduction without full committee approval, and he needs consensus including from Powell before formally announcing changes. Warsh is expected to use his Jackson Hole keynote to publicly make the case for why using the balance sheet to transmit policy is preferable. The strategic logic is that if the long end rises another 100 basis points to 5.5 or 6 percent, it would slow the economy and inflation before the front end needs to move, preserving rate cuts as a future easing tool.

GDP came in at 1.5 percent against a consensus estimate of 2.1 percent, though the majority of the miss was attributable to net exports rather than core domestic demand. Meta fell 8 or 9 percent on the day, reflecting further questioning of hyperscaler capital expenditure buildout. Prior growth tailwinds including the World Cup, big beautiful bill stimulus, AI wealth effect, tariff refunds, and CapEx buildout are all beginning to slow or evaporate, and earnings and growth estimates are expected to come down from recent peaks over the next two to three quarters. A nominal or real recession is described as nearly impossible while the government is running a deficit of 6 percent of GDP annually.

September rate hike odds stood at approximately 55 percent at the time of discussion, and a 5 to 10 percent equity pullback could push those odds back down to around 30 percent, providing cover to skip the hike. The working theory is that talking financial conditions tighter through long-end yield pressure, without actually moving the short end, could produce a stable equilibrium by midterms, as hiking close to midterms was described as politically untenable. The speakers predicted that restrictive real rates and tighter financing conditions would eventually spread beyond the AI trade and hit the broader market, and warned that a rotation away from equal-weight and small-cap economically sensitive stocks, which had been benefiting from dispersion during the AI selloff, could occur if growth slows.