Market Overview | April 21, 2026

Summary

• The market is in a rare macro-micro divergence: oil prices, rates and interest-rate volatility are flashing negative signals on the macro side, while retail is actively piling into leveraged ETFs on the micro side. 3x ETFs—described on air as triple-long products—generated as much as $9B in one-day rebalance buying, while hedge funds in the prime book have been shorting at a three-standard-deviation pace for 2 straight weeks. “When the macro and micro stay split, that is often when the market is relatively dangerous.”
• The oil shock is worse than the Russia-Ukraine war and could be worse than 1973 if the stalemate persists. A 1- to 2-month Strait of Hormuz stalemate could push oil to $120 or even above $150; even an immediate reopening would leave supply-side damage that takes several months to more than half a year to repair. Once strategic reserves and commercial inventories are lower in May-June, true demand destruction begins—“the pain of reduced demand is even more painful.” As a reference point, every 10% rise in oil prices affects GDP by 0.1 and lifts core PCE by +0.04.
• The Fed is split, and room to cut is narrowing. 3 officials shifted toward a more neutral stance at the last meeting; another account says 4 votes are moving more hawkish. Powell is stepping down as chair but staying on as a governor, with a head-down one-liner that Jin read as a silent protest against Trump. Kevin Warsh wants to use median/mean PCE—lower than the current measure after stripping out tariffs—to create room for cuts, but the Fed operates by vote and his ability to whip support is uncertain. The market had expected 2 cuts this year; room to cut may now be down to 1 or fewer. The 10-year is around 4.46%, the 30-year is above 5%, and volatility is rising: “Danger signals are accumulating across multiple dimensions.”
• The micro logic is real and only just beginning to get priced. Compute is directly equivalent to productivity: for the past 20 years, chip iteration mostly made phones better, but now “GPT-5.5 is already more useful than Claude 4.7,” tokens are growing 4x a year, and coding is cheaper than hiring people—“at least you can cut people.” The economy is shifting from a pharma-style model of taxing software back to a factory model, and the semiconductor bull market is “far from over.” The market is overpricing the short term—Jin compared it with calling Intel overpriced at $58—but the long-term value is still nowhere near fully reflected.
• Abundant liquidity is why the market failed to fall last week. Deregulation is putting several hundred billion dollars of additional capital into the banking system; Goldman Sachs estimates the system could absorb another $4T of assets. Term premium minus swap spread in the 10-year is down to roughly 25bp, while SOFR briefly fell below the Fed’s repo rate: “Liquidity between banks has become excessive.” Net issuance is also low ahead of Q2, allowing leveraged positions to build quickly.
• The real correction is more likely at year-end or early next year, will last months, and will be triggered by funding. Once hyperscaler FCF falls, Wall Street will inevitably price in capex cuts; AI investment “will eventually have one miss.” Combined with oil’s 6-month lagged hit to GDP—“just treat it as having collected a tariff”—that could create a synchronized shock. The pullback in optical names and memory should be bought decisively. “You never need to worry about selling too early.”
• Watch Meta’s CDS. By May 1, it had risen by a dozen-plus basis points from its average—“pretty scary,” though not yet at Oracle’s level. If the market prices Meta like Oracle, it will be difficult for Meta to make a Larry-style all-in bet: “Meta is internally a little more chaotic.” Microsoft could also become a casualty. Cloud players have the highest leverage, with real cash flow not arriving until after 2028; semiconductors are sold out through 2028 with a $1.5T backlog, but any wobble in spending will trigger panic.

Deep dive

1. Epic Divergence: Leveraged Retail vs. Three-Sigma Shorts

• Jin opened with a positioning map: CTAs have no buying left, and month-end rebalancing is selling flat. On the other side, retail flows are actively moving into leveraged ETFs. 3x ETFs—described on air as triple-long products—generated as much as $9B in one-day rebalance buying because 3x products automatically chase higher prices; meanwhile, the prime book has been short for 2 straight weeks at a three-standard-deviation pace. “Hedge funds are selling heavily.”
• A review of CTA flows: in the month before March 31, CTAs sold $180B; they then bought $40B the following week as the index surged, $80B the week after as it rose 4-5%, and were forced to buy $20B the week before last to hold their positions. These were all mechanical purchases, indifferent to whether prices rose or fell. “CTAs won’t buy from here.” The forced buying is spent; what remains is leveraged money from retail and long-only funds.
• The next move is unlikely to be a broad selloff; it is more likely to be rising skewness. The indexes may not fall much, but “the number of stocks going up will become very small.” Market breadth has narrowed to semiconductors as the only area doing the lifting. Most of Jin’s office colleagues are still down this year: investors who went all-in on 3x products at the end of March are up 40% YTD, while those who missed the trade have “terrible performance. This trend will continue.”

2. Oil: True Demand Destruction Has Yet to Begin

• This oil shock is “much worse than the previous Russia-Ukraine war, and potentially even worse than 1973.” The 1973 embargo did not last very long, while the current stalemate has no visible solution. Even if the Strait reopens immediately, the baseline oil price will not return to its previous level; supply-side damage will take several months or even more than half a year to repair.
• The current move is only a surface-level price increase: China and the US are releasing strategic reserves and commercial inventories, while demand has not yet fallen. Once stored oil is depleted in May-June, the economy will have to go through actual demand reduction. Jin used memory chips as a comparison: memory is merely getting more expensive now, while last year’s year-end episode was a genuine supply shortage, with memory sticks in Huaqiangbei marked up to several times current prices. “The pain of reduced demand is even more painful.”
• The quantitative anchor is: every 10% rise in oil prices equates to a 0.1 GDP impact, +0.04 core PCE and +0.02 PCE. If the stalemate lasts 1-2 months, oil “could reach $120; I think it could go above $150.” Inflation expectations would then take a real hit, leaving the Fed “in a very difficult position.”

3. A Divided Fed and Dangerous Rate Signals

• The new variable is a crack inside the FOMC: 3 officials shifted toward a more neutral stance, while another account says 4 votes are moving more hawkish. Powell is stepping down as chair but staying on as a governor. When asked what he planned to do next, he said, “I lower my head.” Jin read it as a silent protest against Trump: “Your term isn’t over yet, so I’m still in there making things unpleasant for you.”
• Kevin Warsh wants to work with median/mean PCE: core inflation is lower after excluding tariffs than under the current measure, which would make cuts easier—Trump has repeatedly called the Fed “too late.” But the Fed is a voting institution, and Warsh may not have the pre-meeting vote-whipping ability associated with Bernanke or Yellen. He is also viewed as hawkish and “might symbolically do some QT.” Jin had originally expected 2 cuts this year; he now thinks room to cut may be down to 1 or fewer.
• Rate markets are already flashing red: the 10-year is around 4.46%, near 4.5%; the 30-year has broken above 5%; and 10-year volatility is rising. The front end even priced a hike last night. “But I don’t think a hike is possible.” The right way to buy the dip is to wait until the rates market starts pricing a recession; yesterday was still pricing “continued war,” not a bottom.

4. Compute = Productivity: From Pharma Model Back to Factory Model

• This is the entire foundation of Jin’s micro bullishness. For 20 years, chip upgrades were concentrated in phones and “did not directly become productivity,” unlike a steam engine whose higher output directly produced more clothing. Now every improvement in process technology and packaging translates directly into productivity. The evidence is immediate and experiential: “Try using 5.5 now… GPT-5.5 is already more useful than Claude 4.7.” The underlying driver of the late-March rebound was the move from Opus 4.6 to GPT-5.5, which confirmed for the market that “AI can definitely become productive.”
• Demand is just as real: tokens are growing 4x a year, and coding with GPT-5.5 is “much cheaper than using people—it lets me cut people.” The results are immediately measurable in trading, and “I have no price sensitivity to the price of tokens.” Once people are cut, demand falls and the economy worsens; “that is the story of the next round.”
• Wall Street’s lag is the opportunity. 20 years of a software-taxing model hollowed out the US and European economies, and the market still “really dislikes highly capex-intensive” businesses. Under a factory model, “you simply have to keep buying semiconductors.” There is genuine short-term overpricing—Jin compared it with calling Intel overpriced at $58: “I didn’t sell it—I sold it and then bought it back.” “It’s spring now… and spring can still bring a cold snap.” The long cycle runs for years and is far from over. Jin’s own positions are light; he holds Nokia and AMD and “isn’t likely to make many moves.”

5. Excess Liquidity Explains Why Last Week’s Selloff Never Came

• Goldman Sachs’ FCI shows financial conditions continuing to improve. Risk-based capital and other capital ratios are being adjusted in tandem, putting several hundred billion dollars of additional capital into the banking system. Goldman estimates the system could absorb another $4T of assets. “I have never seen money released through something equivalent to deregulation on this scale.” How much is ultimately released remains to be seen.
• The direct evidence is in the plumbing: the 10-year term premium minus swap spread—the real return from funding long bonds with SOFR—has fallen to just the low 20s bp, while SOFR briefly traded below the Fed’s repo rate. Banks are willing to carry 10-year volatility and margin calls for returns below 25bp, “which shows that bank liquidity has become excessive.” With net issuance low ahead of Q2, “the market should have corrected last week,” but this liquidity prevented it.

6. The Year-End Correction Script: Funding Crisis Meets Macro

• Hyperscalers are ahead of the market in understanding the stakes. They know that “if I don’t build factories today, I’m finished in the AI era”—just as in Monopoly, failing to buy property means losing. Wall Street dislikes capex, so once FCF falls, “it will keep pricing in a need to reduce capex.” Cost discipline is already appearing: Google is running the numbers on TPU v8 at every step, wants to order directly from TSMC but was blocked and had to go through MediaTek, and wants Nvidia to handle packaging rather than TSMC. “CoWoS is too expensive” is an important reason.
• The first crisis path is an expectations reset: the bar keeps rising until “you eventually have one miss,” and shrinking FCF then turns the market back toward questioning semis—even though semiconductor supply is sold out through 2028 and the backlog is $1.5T. The correction in optical names and memory should be a clean one, and Jin’s instruction is to “buy decisively,” because this is a change in the operating model of the past 30 years.
• The second path runs through credit. Meta’s CDS had risen by a dozen-plus basis points from its average by May 1, “pretty scary” but not yet at Oracle’s alarming level. If Meta is priced like Oracle, it will be difficult for the company to make a Larry-style all-in bet. “Zuckerberg is also very decisive, but… Meta is internally a little more chaotic.” Over the next 6 months, Microsoft “could potentially become a casualty.” Cloud is the most levered part of the ecosystem, with real cash flow arriving only after 2028; like insurers or power plants, rising CDS would force them to abandon even a small part of their investment, with major market consequences.
• The timing is initially late this year or early next year, with the correction lasting several months and coinciding with oil’s 6-month lagged hit to GDP—“just treat it as having collected a tariff.” One commodity could become an opportunity instead. The conclusion came down to 2 lines: “You never need to worry about selling too early,” and “the semiconductor bull market is far from over. We have been saying this for more than a year.”