PodBrowser
Forward Guidance

The AI Unwind Is Forcing A Historic Market Rotation | Weekly Roundup

Friday, 17 July 2026 · 4 min read · Listen to the episode ↗

This week's discussion centers on the AI-driven market unwind, which the hosts describe as likely in its early innings, with the S&P 500's roughly 40% Mag 7 and 20% semiconductor concentration framing a bubble they consider larger than the 2000 dot-com episode. Three-month implied correlation remains sub-10, mirroring the summer 2024 yen volatility regime, and momentum unwinds historically require significant positioning to clear before any recovery can be trusted.

One-year forward inflation breakevens are signaling approximately 1% inflation, a level last seen when the Fed was actively easing, and two-year forward breakevens have fallen below 2%. Despite this, Governor Jeffrey Schmidt of St. Louis continues to argue rate hikes should not be taken off the table, and the Fed has now run above its inflation target for 63 months. One speaker argued the core problem is the Fed's reliance on backward-looking, expired data, with dozens of private forecasters consistently outperforming the Fed on inflation and growth projections. Kevin Warsh testified to Congress that Fed members become anchored as hawks or doves and cannot update their views as new information arrives, compounding institutional forecasting errors over time.

Waller's recent speech caused the two-year yield to sell off and hit a new high for the year before a subsequent CPI print reversed the move entirely. One speaker argued that by attempting to suppress volatility through forward guidance, the Fed has instead transferred that volatility into the two-year yield without reducing overall variance. The two-year has front-run the Fed in virtually every instance, driven by market participants independently reading incoming data rather than following Fed signals.

The current AI-driven market unwind was described as likely in its early innings. The S&P 500 is approximately 40% Mag 7 and 20% semiconductors, making it unrepresentative of the broader real economy, and the current concentration was characterized as a bubble larger in magnitude than the 2000 dot-com bubble. All three hosts independently prepared implied correlation charts for the same week without prior coordination, converging on that indicator as a key signal. Three-month implied correlation remains sub-10, similar to the regime seen in summer 2024 around the time of yen volatility. One speaker cautioned that momentum unwinds historically take a long time to digest and require significant positioning to clear before any recovery can be trusted, referencing Renaissance losing approximately 10 to 15% in a single week during a momentum unwind in October of the prior year. The last 30 to 40% of levered momentum moves was attributed to levered retail rather than institutional capital, meaning value buyers still have a long way to go before stepping in.

Credit spreads for hyperscalers were noted to be breaking higher, potentially making it disadvantageous for those companies to raise additional debt. One speaker raised the scenario that if AI models prove cheaper than expected, hyperscaler capital expenditure could decline and trigger a short squeeze in hyperscaler stocks, describing that as the outcome most market participants would least anticipate. Semiconductor and memory companies currently printing 80% margins were expected to face severe competition. The telecom boom was offered as an analog, with infrastructure builders going bankrupt while value was ultimately captured by companies built on top of that infrastructure. A prediction was made that if AI models continue getting cheaper, value businesses could grow margins and passive ETFs could begin chasing beaten-up value stocks, potentially reversing a roughly 20-year imbalance. The growth versus value chart was said to show what appears to be a double bottom tested in 2020 and retested now, though one speaker acknowledged being burned too many times expecting a value rotation that turned out to be only a local top.

The Japanese two-year yield has been grinding higher on a near-daily basis, and the Bank of Korea raised rates, attributed to strong economic returns and resulting inflation. Large asset managers repatriating capital to Japan caused USD/JPY to fall, the yen to strengthen, and Japanese yields to drop sharply. Korea's policy actions were described as potentially laying the groundwork for a Bank of Japan rate hike at the end of the month, and a BoJ hike combined with continued repatriation flows was expected to further strengthen the yen, weaken the dollar, and hit the Nasdaq hardest. Despite large global imbalances in FX and rates, rate volatility and FX volatility remain very low, which one speaker found surprising and concerning. A carry trade unwind was described as capable of producing globally destabilizing margin calls, with potential cracks possibly becoming visible after options expiration. Mortgage rates hit a new high as of the recording date.

On oil and geopolitics, Strait of Hormuz traffic was said to have returned to near zero and global oil inventories were described as more drawn down than during the previous episode of tensions. Ukraine was described as striking Russian oil and gas refining and transportation logistics at a rate not seen across the multi-year conflict, with Russia identified as a top three global supplier of refined products. Despite these pressures, one speaker said oil is more likely to go lower in the short term, noted it is already approximately ten dollars above recent lows, and acknowledged having no clear trade. Trump was characterized as deliberately manufacturing volatility through the Iran conflict in order to control and then suppress it, on the theory that market weakness driven by the AI unwind and earnings cannot be controlled if he did not create it. Credit spreads and the ten-year yield were identified as the key indicators to watch for how badly the Iran situation could affect markets.

This summary was generated from the episode transcript and can contain mistakes.