The AI Trade Is Finally Cracking | Weekly Roundup

The yen carry trade unwind and the momentum factor implosion are structurally intertwined, with flows into AI semiconductors and US tech equities having been reflexively fueled by yen and Korean won weakness against the dollar. The momentum factor saw a four standard deviation move on the unwind, and yen intervention by the Ministry of Finance appeared deliberately timed around the NFP release to maximize market disruption. Stocks in Japan, Korea, and Taiwan in the semiconductor space reversed two to three months of price action in just a few days, behavior the speaker characterizes as bubble-like.

Mag Seven weakness was the initial catalyst, with cracks spreading into semis, AI, Korea, and Taiwan. Amazon announcing potential capex cuts spread that weakness from the Mag Seven into the broader AI trade. Meta considering selling access to its AI compute raised doubts about whether infinite demand for compute actually exists, and a widely circulated tweet claiming a significant memory efficiency architecture breakthrough by a team spun out of OpenAI contributed to the unwind. These catalysts only mattered because of extreme momentum positioning already built up in the AI trade, with triple levered ETFs representing around one hundred billion dollars moving in ways disconnected from AI fundamentals. Approximately 30 percent of hyperscaler income earlier in the year came from markup of AI lab private valuations on their balance sheets, making the AI trade reflexive in both directions. OpenAI delaying its IPO added further uncertainty to the thesis.

The current macro setup is described as oddly similar to July 2024, when the Fed made a hawkish pivot mistake around a cold CPI print. The Fed threw out dots suggesting potentially three hikes this year at the same time labor is rolling over. The jobs report was mixed, with the unemployment rate falling only because labor force participation shrank rather than because of strong demand. Wage growth was described as down in the dumps with no signs of revitalization, and there is no reason for the Fed to hike into a labor market showing job growth around 50 thousand to 60 thousand per month. An LLM analysis of Fed language versus inflation swaps showed Fed sentiment turned hawkish right at the pivot point in July 2024, mirroring current conditions.

ISM PMI data came in weaker with the new orders versus inventories ratio turning downward, and manufacturing PMIs are expected to follow that ratio lower with a two-month lag. Oil is back below 70 dollars and its trajectory contributes to inflation rolling over. Iran war and acyclical factors have elevated core inflation, but all forward-looking indicators show it coming down meaningfully. Central banks that hike into energy supply shocks historically reverse within less than a year, with the ECB cited as an example, and core inflation being hot due to supply shocks does not justify Fed hikes if cyclical components are benign. Going from two hikes priced in to one or zero is only marginally better for liquidity and is not a bullish catalyst comparable to rate cuts.

The speaker described themselves as a rip seller rather than a dip buyer, with a negative bias on NASDAQ and tech and a long bias on gold, viewing real yields as appearing to peak. The next two months were characterized as the most treacherous period going into the back half of the year due to crowded risk-on positioning, with AI bulls and economic reacceleration bulls seen as at risk of unwinding simultaneously as liquidity wanes and the Fed remains hawkish. Record yen shorts, dollar length, and SOFR shorts were described as colliding in a potentially destabilizing way.

Bitcoin held above 50 thousand through the AI unwind, with crypto moving inversely to the AI selloff in semis. The speakers doubt Bitcoin can reach 100 thousand without Michael Saylor returning as a buyer, though it could bounce from current levels simply by removing selling pressure. US deficits remain around 6 percent of GDP with debt continuing to climb and issuance being manipulated, which keeps the structural case for debasement trades intact. Bitcoin and debasement trades are preferred over the NASDAQ as economic growth slows alongside sticky inflation, and every six to nine or twelve months a liquidity risk-off event occurs after which Bitcoin tends to rise as liquidity is added.

Most crypto tokens are described as structurally worthless because they represent no claim on equity even when associated with a named protocol. The Venice token is cited as a specific example, described as a double debit where equity owners hold the token but the token itself confers no equity claim and equity owners can sell it freely. Hyperliquid is offered as a contrasting case, having gained a near-religious following because its founders were perceived as not trying to exploit token holders. The AI trade is identified as where most tourists have migrated from crypto, and the speakers expect some Phoenix-like crypto opportunity to reemerge over the next one to two years, though they caution that the old dynamic of everything in crypto pumping together has largely gone away.