Topics Pioneers Insight Method Author
Back

Market Overview, November 10, 2026

·Shanghao Jin

Rendering the long image…

Topics Trillion-Dollar AI Capex and Bubble Risks Enterprise Private AI and Real Compute Demand Macro Asset Allocation in High-Interest Regimes
Pioneers Shanghao Jin

Summary

  • Shanghao Jin believes last week’s semiconductor selloff mainly reflected tight liquidity and forced fund liquidations. OpenAI and Anthropic’s funding needs squeezed market liquidity, making post-election concerns about construction easier to price in: “When liquidity is scarce, every ghost story gets heard” (“流动性少的时候,所有的鬼故事都可以被人家听得到”). He cited roughly $40B in fund liquidations on Thursday and $45–50B in leveraged ETF rebalancing to explain why funds that were more convinced by AI and more concentrated in the trade suffered the steepest losses.
  • He still expects semiconductors to remain undersupplied next year and the year after, possibly longer. Power, funding, and regulation will affect compute deployment, but they do not mean demand will disappear. Meta could have others build data centers and lease capacity itself; that would not mean it no longer needs compute. He rejects market pricing that assumes flat or declining demand in 2028 and says semiconductors are still a “sell everything you can make” market.
  • AI penetration should be measured against data, not the number of programmers already using the tools. “It’s not people that are being penetrated. AI penetration is about data.” Users moving from Pro Plan to Max Plan and then exceeding their limits shows that usage can keep growing even when user coverage is nearing saturation. He says end-to-end AR has continued to rise since October and argues that $300B of coding demand cannot cap the entire market.
  • His threshold for buying the dip is whether credit losses on the scale of 2008 have occurred. Based on data he recalls reviewing, over the past 30 years, buying at any point after a sector selloff accompanied by forced liquidations and holding for a year made money in every case except 2008. That exception involved impairment of underlying assets, difficulty pricing them, and counterparty risk. He did not buy the dip because he already had a substantial semiconductor position; “not buying the dip” does not mean he has turned bearish.
  • A short-term liquidity shock should ease, but the ceiling on long-term valuations may remain under pressure. The government cannot stop issuing debt, while monetary policy and bank balance sheets may not be able to absorb it at the same pace, splitting the “two-way flywheel.” He thinks this will push term premiums higher and could keep valuations near 20x for an extended period, versus 23–24x in a bull market. What matters more for financing is interest-rate volatility, not a gradual rise in yields.
  • He is much more concerned about banks and financial companies than semiconductors. Easing financial regulation in the US could face greater political resistance, while Europe may wait until pressure becomes severe before supporting the bond market. He worries about a repeat of the pattern in which monetary policy ultimately resorts to money printing, followed by higher taxes and tighter regulation, with lower efficiency as the long-term cost.
  • His personal view is that the US government will shut down next March, but he treats that as a scenario to monitor. A continuing resolution may pass in December, but budget talks could still stall because neither side is willing to compromise. Meanwhile, he expects current liquidity to recover: easing oil prices should first reduce rate volatility, then bring bond buyers back and help credit markets recover. It is like rebuilding a dam and waiting for the reservoir to refill; that takes time.

Deep dive

1. Forced liquidations drove excessive pricing of post-election bearish narratives

  • In the speaker’s account of the election results, Democrats won both chambers and Texas; he believes this intensified market concerns about approvals and construction for AI data centers. Those concerns extended to 2027 and 2028: could today’s severe semiconductor shortage turn into a glut? He thinks these questions were priced in too aggressively because there were too few incremental buyers.
  • His explanation starts outside equities: stress appeared first in rates, bonds, and foreign exchange, then spread to stocks. OpenAI and Anthropic also needed funding, pulling more capital out of the market. “Every ghost story gets heard” (“所有的鬼故事都可以被人家听得到”) when there is not enough money coming in to push back.
  • He estimates that funds with roughly $40B in assets were liquidated on Thursday. A Morgan Stanley report published that morning said leveraged ETFs with $45–50B in assets were due to rebalance, reportedly around 2 p.m. Based on his reasoning, the affected funds needed to finish liquidating before then; the funds most convinced by AI and most exposed to related stocks were hit hardest.

2. AI penetration is about data use, not user coverage

  • He directly rejects the claim that demand has peaked: end-to-end AR has risen since October; OpenAI may not be growing as quickly, but Anthropic is “definitely going up.” He also questions the same researchers who first emphasized the countless applications beyond coding and now use $300B of coding demand as a ceiling.
  • Even if every programmer starts using AI, that does not cap usage. He describes his own path: Pro Plan used to be enough, then he needed Max Plan and began using far more beyond the plan limits. “It’s not people that are being penetrated. AI penetration is about data”; the measure is whether data sitting on servers is actually being used.
  • He calls AI a tool for opening up bandwidth between people and data: using old data requires programming, which generates new data in the process. The railway analogy is how many people it connects, not whether every planned track has been built. By that measure, AI is still at “the beginning of the beginning.”
  • Employment is another way he gauges how deeply AI has penetrated. He says that since Claude Code launched in March, the economy has shed an average of roughly 3,000–4,000 jobs a month, which is not much. Jobs in data-sensitive industries, including IT, government, finance, and manufacturing, have not disappeared quickly; overall job losses remain roughly around the average. The changes in data use have therefore yet to translate fully into productivity gains for people, leaving related demand in its early stages.

3. More constraints on compute do not mean semiconductor demand disappears

  • On rumors such as Meta canceling orders, he says several questions need to be separated: whether a company will build its own compute center, whether it needs compute, who will build it, and what the product is meant to do. A reported cancellation could simply mean someone else builds the facility and Meta leases it. He thinks Meta is very short of CPUs; usage rose after it launched Muse, which may improve the return on investment.
  • Building compute capacity requires CPUs, GPUs, optical modules, and memory, as well as power, funding, and regulation. He acknowledges that the latter three all have an impact, with regulation especially negative. Still, he expects semiconductors to remain in short supply “next year, the year after, and possibly longer”; the added constraints have not brought demand below supply.
  • Memory is the area he has consistently liked less: as power, funding, and other inputs become bottlenecks too, components that once mattered especially may lose some of their premium. But he thinks even memory has fallen too far. A decline in scarcity does not mean it is no longer scarce: “You sell every chip you have.”

4. His threshold for buying the dip is credit losses on the scale of 2008

  • He repeatedly asks: “Do you think this is 2008?” Based on data he recalls reviewing, over the past 30 years, whenever a sector plunged on major negative news and funds were forced to liquidate, buying at any point after the liquidation event and holding for the following year made money—with 2008 the sole exception.
  • The mechanism behind that exception matters: it took more than a year to wind down Lehman; debt and CDOs were difficult to price; the underlying mortgage loans were already at risk of default; and the assets themselves had lost value. Price discovery, counterparty risk, and tightening liquidity became entangled in a credit crunch, so buying after forced liquidations could still lose money.
  • He thinks the current semiconductor setup is the opposite: the sector has been priced too far down. Even after Friday’s and Monday’s rebounds, valuations are still around 2022 levels, the long/short ratio is near a recent low, and semiconductor companies’ earnings are growing quickly. He cites Sivers and POET: POET has not yet received orders directly and is still under review; Sivers is valued at about 3x cash, and every chip it produces is sold.
  • His own position sets the boundary: he already has a substantial semiconductor position, so he did not keep buying the dip. For those with lighter positions and lower risk, he says the question to answer before adding risk is whether this is 2008.

5. Fiscal issuance cannot stop, while monetary support may struggle to keep up

  • He uses the “two-way flywheel” to explain past liquidity expansion: government spending funded by debt issuance needs balance-sheet capacity on the other side to buy the bonds. Japan’s YCC, central-bank QE, and eSLR, which allowed banks to increase leverage, took different routes but all answered the same question: who will buy the debt? Fiscal spending is tied to votes, so neither party can easily stop it once in power.
  • Political resistance to monetary support is growing. Energy prices, diesel crack spreads, and crude prices shape how people experience inflation; even without a wage-price spiral, the higher cost of living caused by rising oil prices can affect votes. He thinks Elizabeth Warren will chair the Senate Banking Committee next session and worries she will put greater pressure on eSLR and related policies.
  • He describes the Fed’s path as managing only the short end, leaving the long end to the market, then freeing up private balance sheets through deregulation. He says Kevin Warsh calls QE “reverse Robin Hood.” The problem is that Warsh does not want to print money to buy long bonds, while a Democratic Congress may block financial and regulatory reform; both routes to expanding liquidity could be constrained at the same time.
  • He therefore expects term premiums and swap spreads to trend higher, but gives no target for Treasury yields. For credit markets, interest-rate volatility matters more: real-money investors will still buy bonds when yields rise steadily, while sharp swings scare buyers away. That does not necessarily mean AI funding is in poor shape.

6. Europe may repeat the pattern of delaying support, then tightening controls

  • He thinks the risk in French government bonds has not gone away despite a recent decline of about 20 basis points, and says no policy to improve the fiscal position is possible before the French presidential election. The ECB could activate TPI but will not; in his account, its sole mandate is to control inflation, it raised rates in September, and Europe will not ease regulation.
  • His reference point is 2012: austerity failed to truly constrain fiscal spending, monetary policy initially did not cooperate, and in the end money still had to be printed. More regulation and taxes followed, bringing a deeper long-term efficiency cost. He worries the same political logic will recur. He is more concerned about risk in banks and financial companies, while remaining very bullish on semiconductors.

7. Shutdown risk and liquidity recovery operate on different timelines

  • His personal view is: “The US government will shut down next March.” A CR may pass first in December, but after Democrats won by mobilizing opposition to Trump, budget compromises could become harder, and the president could veto more aggressive policies. Republican concessions could change the outcome, but he thinks Republicans “are not likely to compromise much either.”
  • He stresses that this is “scenario analysis” and that there are many variables along the way; long-term political risk does not directly translate into a signal to liquidate tomorrow. Even if the system really broke and was ultimately rescued with money printing, the follow-on regulation would take years. He cites the regulatory process after the 2008 crash, which was not completed until 2014.
  • The current episode, by contrast, is a liquidity shock that will pass. He describes credit and rates markets as stabilizing, not easing; credit markets are still weak, but investment-grade issuance will decline starting in December, which should help. The chain from the Iran issue and easing oil prices also takes time: rate volatility slows, buyers return to bonds, credit markets recover, and liquidity comes back.
  • His analogy is that after a reservoir collapses, “rebuilding the dam” still leaves the reservoir to fill gradually. Short-term liquidity “will definitely recover,” but long-term valuations may remain under pressure: a bull market that once supported 23–24x may only sustain around 20x. The market can trade in a range, but he thinks it will be difficult for short positions to stay this elevated for long.