← All episodes · Forward Guidance
The AI Trade Is Finally Cracking | Weekly Roundup
Overview
In a stripped-down holiday-week episode (just the two hosts, with Tyler out on a road trip), the pair dig into a violent "momentum factor implosion" that hit markets that day — a four-standard-deviation unwind in momentum factors that coincided, as it so often does, with a Yen intervention from Japan's Ministry of Finance. They trace how a fragile, highly reflexive AI/semiconductor/memory-DRAM trade got knocked over by two relatively minor news catalysts: Meta reportedly considering selling excess AI compute, and an anonymous but reputable account tweeting about a memory-efficiency architecture breakthrough coming from an ex-OpenAI team. Neither headline needed to be true — positioning was so crowded and levered (with an estimated $100B+ sitting in triple-leveraged tech ETFs) that any narrative crack was enough to trigger a cascade through Mag Seven, semis, and Asian markets (Korea, Taiwan, Japan).
The conversation then pivots to the macro setup, which the hosts argue rhymes closely with July 2024: a Fed that just turned hawkish (dot plots showing potential hikes) right as growth and inflation data are actually peaking and starting to roll over, mirrored by a weakening jobs report, falling wage growth, and a shrinking labor force participation rate. They walk through the jobs report, ISM PMI internals (new orders vs. inventories rolling over), and argue the Fed has essentially no case for hiking given a labor market that is weak, not overheating. They spend significant time steelmanning the hawkish case (wealth-effect-driven inflation, administration equity-market manipulation propping up jobs) before concluding the dovish case is much stronger.
The episode closes with portfolio positioning: fading Mag Seven/Nasdaq strength via selling bounces, preferring gold, silver, and the broader debasement trade, a wary but improving view on Bitcoin/MicroStrategy (Sailor) as forced-seller pressure fades, and a skeptical tour of crypto's token/equity structuring problems (using the Venice/Voorhees situation as an example), plus lighter banter about the World Cup.
Topics
The momentum factor implosion and Yen intervention
The episode opens with the hosts describing a four-standard-deviation "momentum factor" unwind happening that day — a total factor implosion that, per one host's pattern recognition, always seems to coincide with something breaking in the Yen. Overnight, Japan's Ministry of Finance appeared to finally intervene to support the Yen, something they'd been expecting for a while. One host mentions an as-yet-unreleased interview with FX strategist Brent Donnelly, who predicted the MoF would time an intervention around the NFP jobs print because they like a catalyst to generate real movement.
The other host frames this as classic reflexivity: nobody notices a factor unwind while it's quietly building (the Yen weakening against the dollar, the Korean won weakening, huge flows into AI/semiconductor/US tech), but everyone notices when it violently reverses, because the same correlated, intertwined flows that pumped the trade up unwind together on the way down.
Cracks forming under the AI trade's supports
The hosts catalog a string of small "cracks" showing that the forces that had been pumping risk assets higher are losing potency: the US rescinding the Fable model export ban on Anthropic produced no market reaction; Trump has been tweeting repeatedly about Micron (allegedly because he's long and frustrated his "manipulation tweets" aren't working anymore); and the Magnificent Seven has been quietly underperforming for weeks. The bull case had been that Mag Seven was cheap and would soon retake leadership and push the Nasdaq to new highs — but the hosts argue the more likely dynamic is the opposite: Mag Seven weakness is the leading indicator, as companies (citing Amazon's announcement) start cutting CapEx to protect cash flow and reduce leverage, and that weakness then spreads outward into semis, AI names, Korea, and Taiwan.
They explicitly compare the current setup to July 24, 2024: a hawkish Fed pivot (dot plots implying up to three hikes) arriving at exactly the moment labor and inflation data were rolling over, which back then reversed into a Jackson Hole dovish pivot and cuts. One host says he'd be "very cautious" holding a lot of tech risk right now — not because of an imminent collapse, but because the marginal easing of hawkish pricing (two hikes priced in moving to one or zero) is only marginally supportive of liquidity, while broader liquidity conditions look poor.
The two catalysts that triggered the unwind: Meta compute and the memory-efficiency tweet
One host emphasizes that catalysts only matter when positioning is already stretched, and points to the extreme, "outright absurd" outperformance of the momentum factor (largely tied to memory/DRAM/AI trades) as the fuel. Against that backdrop, two relatively small news items were enough to break the trade: first, reports that Meta is considering selling its excess AI compute, which undercuts the "infinite demand for compute" narrative regardless of whether it's actually true — what matters is that positioning was primed to second-guess itself. Second, a widely-viewed (2 million views) tweet from a reputable, semi-anonymous account claiming a major memory-efficiency architecture breakthrough is coming from a team spun out of OpenAI, which fed directly into anxiety about memory/DRAM company valuations.
The hosts describe this as the market's tendency to overshoot in both directions before settling — shaking out both bulls and bears rather than sitting in equilibrium. They note the sheer scale of leveraged positioning (roughly $100 billion in triple-leveraged ETFs tied to these trades) is completely disconnected from fundamentals, meaning bears kept getting blown out on the way up, and now the unwind is violent because everyone was leaning the same way.
Economic data: peak growth, peak inflation, and a weak jobs report
One host walks through how the economic narrative evolved: through 2024-2025 the fear was that AI capex was the sole driver of growth, but the "One Big Beautiful Bill" (OBB) immediate CapEx expensing and tax-refund rebates supported consumers in H1, hiring and manufacturing surveys stabilized, and the growth story broadened beyond AI into small caps, the Dow, industrials, and CapEx-driven names like John Deere and power/build-out plays. But now, alongside cracks in the AI trade, the data suggests the economy is hitting a "crescendo" of peak growth and peak inflation simultaneously.
They reference the ISM PMI report released the prior day, which came in weaker, with the leading new-orders-versus-inventories spread turning down — a signal that manufacturing PMIs (which lag by about two months) are likely to follow lower. Wage growth is also described as "down in the dumps." On the jobs report released that day, the unemployment rate ticked down (bucking the broader softening trend) even as there was a fairly significant miss on nonfarm payrolls (NFP). The hosts read this divergence as labor force participation shrinking rather than a healthy labor market — people leaving the workforce, not being pulled in by strong opportunity — which they argue is not a hawkish signal despite the lower unemployment rate.
Why the Fed has no case to hike right now
One host argues forcefully that the Fed has essentially no justification for hiking: wage growth shows zero signs of the kind of wage-price spiral that existed in late 2021, the labor force participation drop reflects supply constraints rather than a labor market genuinely overheating, and hiking into an economy already producing only 50-60K jobs a month (tepid, "trembling along") would be backwards — central banks hike to cool an overheating economy, not to further suppress an already-soft one. He notes the Iran war and other acyclical shocks pushed core inflation up temporarily, but all forward-looking indicators point to inflation coming down meaningfully, especially if equity markets stall (which feeds reflexively into the real economy).
They reference an Andreas Steno Larsen chart comparing inflation swaps to an LLM-derived measure of FOMC sentiment, showing the Fed turning hawkish at almost exactly the wrong moment — mirroring June 2024. One host jokes that history is repeating so obviously he wonders what he's missing, prompting both to steelman the hawkish/devil's-advocate case: a wealth effect from AI-driven equity gains (potentially aided by the administration propping up markets, e.g., by taking equity stakes in AI companies) could keep the labor market tight even as core inflation, supported by cheap oil, refuses to fall. But they conclude the Fed can only really influence the cyclical components of inflation, not acyclical supply shocks, and that a hawkish committee composition would need to see a genuine step-function improvement in the labor market to justify holding the line — something they don't expect given a fading fiscal impulse (tax rebates rolling off, capex cooling) and a possible give-back of an Iran-war-driven pull-forward in demand, similar to the tariff pull-forward pattern.
Fed communication, Warsh, and rate-cut pricing for the July meeting
The hosts discuss the Fed chair's (Warsh's) recent comments, noting that even without giving explicit forward guidance, saying inflation risks are falling, Iran risks are falling, and inflation swaps are falling is itself a form of forward guidance. They argue it's inconsistent for inflation swaps and forward-looking indicators to be falling sharply while the Fed still signals hikes "because they said so" via the dot plot — the whole point of removing the dots is to let markets react to incoming data rather than a convoluted, reflexive dot-plot signaling game.
On positioning, they note markets are still not pricing 100% odds of a hold at the July meeting, with some marginal probability of a hike still priced in, which they expect to get rectified given recent data and comments. They flag the classic trader mistake of seeing what should happen over the next few months, going all-in too early, getting stopped out, and then watching the thesis play out anyway. They also observe that Fed speeches have dropped sharply in frequency (from 5-10 a week to almost none), which they think is actually healthier after a very talkative period, giving markets time to adjust. Separately, they discuss Warsh's structural view that the real problem is the suppression of the long end of the yield curve (not front-end policy), and that tightening via letting the long end price properly and reducing balance-sheet duration would be a Warsh-aligned way to tighten that's distinct from raising the policy rate.
Positioning: fading the Mag Seven, hurt-locker hyperscalers, and the reflexive AI-lab valuation loop
Asked how they're playing the factor washout, one host says he's more of a "rip seller" than a dip buyer. He argues the Magnificent Seven are stuck in the "hurt locker" for a while even with a relief bounce, because roughly 30% of hyperscaler income earlier in the year came from marking up private AI-lab valuations sitting on their balance sheets — a reflexive loop where if the AI "receivers" trade unwinds, it's also bad for the hyperscaler "payers." He's selling bounces in mag names, notes SpaceX's upcoming QQQ inclusion date is drawing some selling to make room, holds a negative bias on Nasdaq/tech, a long bias on gold and silver (the "fading the hawkishness" trades), and thinks the next two months are the most treacherous stretch of the year given how crowded the risk-on view still is.
The other host says he wants to buy the dip but doesn't think it's time yet, and describes a personal framework for trading these "parabolic rotation" cycles: decide whether you're investing or trading, decide your time horizon, and have a pre-set plan (e.g., buying in the "second or third inning," selling some in the "seventh inning," holding a small residual position through a possible blow-off top) so you're not reacting emotionally when a single tweet moves the tape. He still believes in the multi-year Jevons paradox thesis — that efficiency gains (like the rumored Fable/memory breakthrough) increase rather than destroy compute demand, citing the shift since January into an "agentic world" needing far more inference compute than expected — but says the market is currently in a phase of "questioning the thesis," with open questions like why Meta is renting out excess compute and why OpenAI delayed its IPO that will take time to resolve.
Bitcoin, MicroStrategy/Sailor, and the beach-ball-underwater framework
The hosts note Bitcoin and crypto have moved inversely to the semis unwind, having already been heavily bearish and "faded" in prior weeks. They recall flagging last week that people should temper bearishness on the Sailor/MicroStrategy complex. One host describes a mental model of these markets as "beach balls underwater" in both directions: Bitcoin held above $58K despite one large liquidation wick (someone reportedly liquidated for about $700 million, with a brief 50-cent-on-the-dollar wick before recovering), and if MicroStrategy/Sailor's forced buying/selling pressure was what was holding the ball underwater, simply removing that hand from the ball is enough to produce a bounce — you don't need Sailor to resume buying, just for the selling pressure to fade. He's skeptical Bitcoin gets back to new highs (100K+) without Sailor as an active buyer, but thinks the near-term bounce is achievable.
The other host agrees, noting sentiment on Sailor/Strategy has been the "laughing stock" for about a year (coming up on the anniversary of Saylor's "free money printer" strategy), and that there's a "shot clock" on being the last bear given the risk of a squeeze. They pivot to Warsh potentially having an "I told you so" moment given markets had misread his press conference as hawkish when it was really just him relaying committee dot-plot sentiment, and reiterate the structural debasement case for gold and Bitcoin: 6% of GDP deficits, still-climbing debt, and continued issuance manipulation.
Crypto's token/equity structuring problems and the Venice/Voorhees example
The hosts discuss where value actually accrues in crypto versus where the industry narrative points, using the newly launched stablecoin landscape as an example (one host says he doesn't want to be Circle in that scenario) and reiterating the case for Bitcoin as the asset that reliably benefits every 6-12 months when a liquidity risk-off event forces central banks to add more liquidity.
They then discuss the Venice situation (tied to Erik Voorhees) as an example of a broader structural problem: a token that has the same brand name as an equity but represents no actual claim on that equity, allowing insiders to "double dip" by owning both the equity and the token while the token itself confers no real economic rights. They argue much of crypto has this issue — tokens or governance instruments that represent "nothing" despite sharing branding with a real protocol or foundation — and that this needs to be washed out before the industry can mature. One host jokes about the public's tendency to be shocked by each new scam when scams are actually the norm, and both agree that projects like Hyperliquid stand out precisely because they don't try to extract value from holders, and that capital will eventually flow toward properly structured assets even if the process of "rinsing out" the bad actors takes time.
Visuals
| July 2024 | Today (July 2026) | |
|---|---|---|
| Fed stance | Hawkish pivot, dots showed hikes | Hawkish dots again, hikes floated |
| Inflation/growth trend | Cool CPI print, Fed stayed hawkish anyway | Inflation swaps falling, PMIs/labor softening |
| Yen/carry trade | Yen carry trade unwind | Yen intervention by MoF |
| Eventual outcome | Reversed at Jackson Hole, then cut | Hosts expect a similar dovish reversal |
Takeaways
- A four-sigma momentum factor unwind hit tech/AI/semis after two relatively minor catalysts (Meta considering selling excess AI compute; a viral tweet about a memory-efficiency breakthrough from an ex-OpenAI team) cracked extremely crowded, levered positioning (~$100B in triple-leveraged tech ETFs).
- The setup is being compared directly to July 2024: a Fed turning hawkish on dot plots right as growth, inflation, and labor data are actually peaking and rolling over — a pattern that in 2024 reversed into a dovish pivot at Jackson Hole.
- Falling labor force participation, not labor market strength, is driving the lower unemployment rate; combined with dead wage growth, the hosts see no case for the Fed to hike, and expect the next 1-2 inflation prints to support a dovish shift.
- One host is a 'rip seller, not dip buyer' on Mag Seven/Nasdaq, citing the reflexive 30%-of-income hyperscaler exposure to marked-up AI-lab private valuations; he's long gold and silver as the 'fading the hawkishness' trade and expects the next two months to be the most treacherous stretch of the year.
- Bitcoin/MicroStrategy: a bounce back toward recent highs doesn't require Sailor to resume buying, just for forced-selling pressure to fade (the 'beach ball underwater' framework) — but a run to new highs (100K+) without Sailor as an active buyer looks unlikely soon.
- Crypto's structural problem is tokens (e.g., the Venice/Voorhees situation) that share branding with an equity but carry no actual economic claim, letting insiders double-dip; this needs to be washed out before capital can flow to properly structured protocols like Hyperliquid.
Notable quotes
There's a lot of cracks forming where the things that were pumping this to the upside aren't really working anymore.— Host (Quinn)
These two headlines, and suddenly you just have this like momentum factor implosion that we're seeing.— Host
I would be very cautious here if I was loaded to the gills with tech risk.— Host (Quinn)
You're reaching this crescendo — what I think is peak growth and peak inflation for the economy... There's just no reason for the Fed to act in either direction right now because of the labor market.— Host (Quinn)
I'm more of a rip seller than a dip buyer.— Host (Quinn)
You have to have a plan going into these moments so that you don't get freaked out in the moment. Because if you're reacting now, you've already messed up.— Host
It's the opposite — when you're surprised when a non-scam comes along.— Host