Market Overview — April 14, 2026

Summary

• The core conclusion: the market remains decisively bullish over the next 2 weeks. The liquidity inflection point was last week, with conditions most abundant this week; the likely turn comes “roughly the week after next”—highly fact-dependent from there, given the U.S.-Iran situation around the Strait of Hormuz and the tax-season liquidity drain—but “uncertainty is not particularly high” this week.
• Over the past month, CTAs have reversed course and sold $180B, breaking the S&P 6700/6600 levels; combined with the highest-ever prime-book short Z-score as of April 9, that still failed to produce the expected 10-15% correction to 6300. There are 4 reasons: large-cap valuations are already below last April’s levels—Nvidia’s earnings are up 70%, yet its stock is effectively back at April’s $90 price; AI downside expectations, including the view that hyperscaler free cash flow will not support enough capex, are fully priced; token demand is rising exponentially; and the economic fundamentals remain relatively healthy. “These opposing forces are why the market has struggled to fall.”
• CTA buying is highly certain: according to the “dogs” data, even a 2.5-standard-deviation drop in U.S. equities this week would still trigger $21B of CTA buying, while a flat market would bring $32B; globally, a flat market implies about $80B of buying, rising to $100-120B over the following month. Last week’s 7% rebound was mainly prime-book deleveraging—CTAs bought only about $10B—while long-only funds have bought just “3 out of 10” and have yet to enter in force, leaving plenty of fuel.
• A breakdown in the Iran talks does not matter. “Whether it negotiates successfully or not… it goes up either way, because the buying pressure behind it is so certain.” The IRGC wants permanent rule and cannot back down publicly, but does not want a real war; Trump’s approval rating fell 4 percentage points after the war began, Polymarket has flipped to a Republican loss in both chambers, and he says a ceasefire must come in 2 months and pass Congress—“passing Congress is impossible”—while also saying the war must end within 3 months. The optimal outcome for both sides is to “insult each other, then let the matter fade away,” followed by a gradual de-escalation.
• The liquidity signals remain strongly supportive. The rate wall fell by 2 standard deviations, SOFR briefly traded below the fed funds rate last week, and about $15B flows back into the RRP reservoir each day, leaving interbank liquidity “too abundant.” The end of tax season in the week after next could drain more than $400B in bank reserves; if the market rallies too far over the next 2 weeks, the near-term adjustment likely comes late next week or the week after. If Kevin Warsh takes over, “his mandate will certainly be to cut rates,” implying an easy policy bias.
• This year’s trade is to find the AI losers. Last year, anything tagged AI rallied first; this year, stocks get hit first, followed by the question: “So what if your earnings are good—can you prove it?” Software fundamentals are not necessarily broken—Cloudflare is the standard example—but AI can uncover “bugs hidden for decades,” calling the security of human-written code into question. Any investment should be directly linked to rising compute demand.
• Two names stand out. AMD’s cards are the weakest of the 3, with growth and margins well below Nvidia’s, but Nvidia cards and TPUs are sold out: “if you can buy a card, put it in no matter what.” AMD is a forced choice with high upside elasticity. Nokia’s long-haul optical-transmission orders are jumping from $1B last year to $10B this year, a market that has not priced it in; tenders likely begin in April, with Google first and the process roughly complete by August, potentially split evenly between Ciena and Nokia. Longer term, “AI run[?]” could put inference into base stations: “I used to be a real-estate underwriter; now I’m a landlord.”

Deep dive

1. Selling Couldn’t Move the Market

• Since year-end, the market had “run too far, too fast,” triggering a reversal in CTA positioning and roughly $180B of selling over the past month. The U.S. accounted for half, or about $80-90B. That selling broke through the S&P’s 6700 and 6600 levels; at the same time, the prime book’s short Z-score as of April 9 hit an all-time high. “Shorts are extremely crowded,” and CTAs kept adding fuel.
• But the decline was far smaller than the speaker expected around the Lunar New Year: a 10-15% correction, at minimum to 6300. The first reason is that valuations are already very low. Apart from Google and Tesla, the valuations of Microsoft, Nvidia, Amazon, Meta and other megacaps are below last April’s levels—earnings have risen, but the stocks have not. Nvidia’s earnings are up 70%, while its stock is effectively back at April’s $90 level, leaving little room for further downside.
• The second reason is that AI downside expectations are fully priced. Wall Street’s assumption is that hyperscaler free cash flow will shrink and leave them unable to fund sufficient capex. The speaker disagrees: financing channels have not even opened up yet, and the money can come from headcount cuts. “People are the biggest drag on productivity,” and “Oracle is already doing these things.”
• Third, token demand is rising exponentially and compute demand is growing 4x a year. The clearest evidence is the recent experience of cloud users: “If you use the cloud frequently, you’ll notice it’s been lagging lately… some people say the cloud has degraded.” In fact, “the cloud has run out of compute.” His framework is that the internet and software era followed a “pharmaceutical model”—build once, then copy and sell indefinitely—whereas AI follows a “factory model”: “Electricity goes in, tokens come out, and you pay every time you talk to it.” The entire sector is shifting from light-asset to heavy-asset economics, but companies producing these “heavy-asset means of production” have not received enough valuation. Storage is the same trade: storage makers are shifting from serving phone and auto manufacturers to serving only compute centers—“go wherever you want.”

2. The Rebound Was Driven by Buyers

• Last week’s 7% gain, a historic-high weekly return, was not driven by CTAs: U.S. CTAs bought only about $10B and global CTAs about $15B. The prime book went through large-scale deleveraging. The rebound also cannot be attributed to U.S.-Iran talks: “Whether the negotiations go well or badly, including whether the weekend talks succeed, it goes up either way, because the buying pressure behind it is so certain.”
• Looking ahead, the “dogs” data shows that U.S. CTAs would still buy $21B even after a 2.5-standard-deviation decline this week; a flat market would bring $32B of buying, while a normal global market implies about $80B. Over the month following April 9, the flat-market case implies $100-120B: “It is unlikely to stay flat, so the probability is high that it buys even more.”
• Fundamental long-only funds bought only “3 out of 10” last week and are still waiting for clarity on the Strait of Hormuz. That potential buying has not yet been released.

3. Politics Constrains the U.S.-Iran War

• On the Iranian side, the IRGC’s objective is permanent rule. It controls more than 40% of the country’s fiscal and oil resources, with “the military doing business” and “self-enrichment” making it extremely wealthy; its children are in the U.S. After the last domestic massacre, public sentiment turned against it, so it needs the broader cause of “anti-Americanism.” It will not back down publicly, but it also does not want to continue fighting the U.S. for real.
• On the U.S. side, Trump’s approval rating fell 4 percentage points after the war began, and the MAGA base is eroding. Polymarket has flipped since last month to show Republicans potentially losing both chambers. Trump has said the war must stop in 2 months and that the outcome must pass Congress—“passing Congress is impossible”—then said it actually must end within 3 months. If he wages an unconstitutional war, “all the responsibility is yours.” With the midterms approaching, he has a strong incentive to end the war. Iran understands this and will not make further concessions at the negotiating table.
• He considers extreme escalation unlikely, whether the U.S. bombs power stations and lands ground troops or Iran strikes Kuwait’s desalination facilities to force Europe into the war. “The best outcome is for everyone to insult each other, then let the matter fade away,” followed by a gradual de-escalation. Before Vance announced over the weekend that the talks had broken down, “not one person genuinely thought this time would end with a handshake… and 40 years of hostility would disappear in one go.” But pressure is building: if the U.S. blockades the Strait, no Iranian oil can get out and China cannot import oil. “Chinese oil tankers already turned around this morning.” Iran’s economy would be near collapse, and “everyone just holds out.”

4. Liquidity Remains Excessive

• On rates, the biggest negative for equities is not the level of rates but rate volatility. The rate wall fell by 2 standard deviations last week, which is itself substantial support.
• On funding, SOFR briefly fell below the fed funds rate last Thursday, showing that interbank liquidity is “extremely abundant.” Money-market funds are putting cash back into the RRP to earn free money, with about $15B flowing into this “liquidity reservoir” every day. The reservoir accumulated more than $2T during the pandemic and bled down for 1.5 years. With CTAs buying, prime brokers unwinding, long-only funds still waiting and the rate wall falling, “we can ignore what happens with the Iran talks.”
• The inflection point comes when U.S. tax payments begin late next week: cash moves from commercial banks through the Treasury into the Fed, draining liquidity. By the end of tax season, bank reserves could fall by more than $400B. The negative impact “may not be very large” because liquidity is currently excessive, but if the market rallies too far over the next 2 weeks, that is where the adjustment would come. On the Fed, he maintains his view that rates will fall this year: core CPI is not facing a wage spiral, and a 10% rise in oil prices affects core CPI by only about 0.04. If Kevin Warsh takes over, “his mandate will certainly be to cut rates.” The market expects QT and a release of the pressure, but he will still cut rates.

5. Software Gets Sold

• Hedge-fund shorting of U.S. software is obvious, but “software fundamentals themselves have no problem,” and the slowdown is “not very related to AI right now.” It simply coincided with the AI narrative. Cloudflare is the standard example: demand is strong—“we run high-frequency trading ourselves; if it goes through the AWS cloud, of course we use Cloudflare”—but the stock plunged because Anthropic said a new model could pose a security risk. AI can search massive volumes of data and find “bugs hidden for decades.” The new model, which he called “Mithius[?],” was trained on B-cards; even Nvidia’s “Well Rubin[?]” has not yet been used. As successive generations stack up, “the security of things hand-written by humans in front of machines is terrifying.” “Take 360 Security, for example—it becomes garbage.”
• His summary of this year’s trading regime: last year was about finding AI winners—“tag it AI, and whether earnings are good or not, give it a rally first.” This year is about finding AI losers—“hit it first, then even if your earnings are good, so what? Can you prove it? Take NET: can you prove that you’ll still be this strong in security 3 or 5 years from now? You can’t, so get hit first.” The conclusion: any investment should be directly tied to rising compute demand.

6. Compute Names Start to Benefit

• He does not gloss over AMD’s awkward position. Its cards are the weakest of the 3; even if it gives away stock, its pricing and margins are below Nvidia’s, and its growth also lags. Nvidia’s 70% growth and 75% margin are well out of reach, while the gap in rack-level packaging is visible. Nvidia has also bought up the HBM supply, so AMD “is not a competitor.” But token demand is enormous, Nvidia and TPU capacity is sold out, the U.S. is not short of power, and construction capacity far exceeds available cards. “The best thing to do now is put in any card you can buy, no matter what.” AMD is the forced choice and a high-elasticity name.
• Nokia combines a short-term tender market with no pricing in the stock and a long-term “AI run[?].” As compute-center interconnect demand surges in the AI era—“a small road between cities used to be enough; now we’re building highways”—long-haul optical-transmission orders on a 5-year cycle are jumping from $1B last year to $10B this year. The only contractors in the market are Ciena and Nokia. Tenders likely begin in April, Google starts first, and the process should finish around August, with Ciena and Nokia potentially taking half each. Longer term, 80-90% of base-station power may sit idle, and “AI run[?]” could put inference into local base stations; Nvidia’s investment in Nokia is for this purpose. “I used to be a real-estate underwriter; now I’m a landlord… I’ve become Hilton.” This is a game changer. “Don’t look down on Nokia.” He may discuss 2 more names next week; interested listeners can DM him or James.

Verification Notes

• Raw captions use “狗们” for the data source, “cloud” for the service, “Mithius” and “Well Rubin” for names, and “AI run” for the term; the intended identities and term remain unconfirmed.