Market Overview — June 23, 2026

Summary

• Contrarian view: long-end rates are entering a long bear market. The market reads Kevin Warsh’s hawkish stance as “control inflation → push down long-end rates → lower U.S. financing costs.” Jin sees the exact opposite: after the Fed abandons the dot plot and forward guidance and “leaves rates to the market,” 5-year and 10-year yields should trend higher rather than lower over a prolonged cycle, steepening the curve and lifting term premium—“Sisyphus is pushing the boulder; rates have to keep climbing because issuance keeps increasing.”
• The dollar’s reserve-currency status is not threatened by the size of U.S. debt. Its reserve status comes from the U.S. being the world’s largest debt-trading market—its depth—combined with the United States’ legal system, English and case law as the financial system’s “source code.” Collateral, CSAs and settlement all run on dollars, so the U.S. can “issue as much as it wants.” The real problem is r>g: when rates exceed growth, public debt crowds out private-sector credit; the larger the government’s claim on resources, the lower the economy’s efficiency—what Bill Gross apparently called Credit Supernova.
• Government borrowing is an addictive drug. “To generate the same dollar of GDP, you need more and more credit growth—it’s a bit like taking drugs.” China’s supply-side reform and restrictions on local-government borrowing did not work: “Money and power are both good things, and they’re hard to quit.” That is why he does not think Warsh’s attempt to turn the clock back to 2008 will succeed; the U.S. has at least reached China’s 2013–14 position.
• The Fed’s final choice is between giving up control of the long end and allowing inflation to rise. Bank deleveraging—from leverage ratios of 35–40 at the time to 15—has no room left, so the central bank must either abandon control of long-end rates or let inflation run higher: “This is the Fed’s dilemma, and the final policy choice at this point.” This year’s bank deregulation could release roughly $4.5T–$5T of deployable balance-sheet capacity to buy Treasuries, temporarily masking pressure from long-end supply and demand.
• The White House cares so much about the stock market because capital markets serve as the reservoir. China’s reservoir is real estate; the entire r>g game only works if growth holds up, and growth is now pinned on the AI productivity revolution. “If the market crashes, you immediately enter a debt supernova.” He believes the risks in the capital markets ahead remain considerable.
• July’s second-quarter earnings are a bigger risk than opportunity for several major cloud providers, with Google the worst hit. The focus is shifting from CapEx to ROI: this year’s CapEx is about $770B, while valuations are already priced through 2028 on a forward basis—“$1T next year, $1.5T the year after; where does the money come from?” Guidance that is too high could crash the market; guidance that is too low could trigger another semiconductor selloff. “You have to get it exactly right.” Google’s front- and back-end books will be dissected, and “Google’s models still aren’t very good this year.”
• Semiconductors themselves have no problem whatsoever. Every node from advanced to mature is in shortage, and GPUs, CPUs, memory, optical modules and MLCCs are all beating estimates; “individual names are fine.” The real risk is in cloud-company earnings above, not in the stocks themselves.

Deep dive

1. Contrarian view: long-end rates are headed higher and the curve will steepen

• First, a recap of the Fed’s decision: rates were unchanged, but there were 2 key moves—less communication and no dot plot or forward guidance, signaling a desire to “leave rates to the market.” The immediate market reaction was a sharp rise in short-end rates and a flattening of the entire curve.
• The consensus narrative is that Warsh is hawkish in order to control inflation, drive down long-end rates and lower the government’s financing costs, because the debt is “not very sustainable.” Jin takes the opposite view: “I see long-end rates differently from most people. Long-end rates should trend higher over a relatively long cycle, the entire curve should steepen, and term premium should rise.” That is especially true for 5- to 10-year maturities, “because demand for 5- to 10-year paper is not as strong.”

2. The dollar’s status and Credit Supernova: the real problem is r>g

• Could debt expansion undermine the dollar’s status? His answer is no. The dollar’s reserve status comes from the United States being “the world’s largest debt-trading market”—“the more debt it owes and the better the liquidity, the more it becomes a reserve”—plus the financial “source code” of U.S. law, English and case law. Collateral, CSAs and settlement are all dollar-based, so “it is very difficult for any currency to replace it as the reserve currency.” The U.S. can therefore “issue as much as it wants.”
• The real pathology is r>g—what Bill Gross apparently called Credit Supernova. When rates exceed growth, the public sector crowds out private-sector credit: “The U.S. government can borrow at 5% or 6%, but the private sector has to look at returns.” Economic growth is fundamentally credit growth; the larger the share of credit taken by the government, “the efficiency of the country as a whole inevitably declines, without question.”
• The most vivid analogy: “To generate the same dollar of GDP, you need more and more credit growth. It’s a bit like taking drugs—you need more drugs to get the same high.” China went down this road much earlier than the U.S., attempting supply-side reform and restrictions on local-government borrowing. “It didn’t work,” because “money and power are both good things, and they’re hard to quit.” U.K. debt has already begun to exceed nominal GDP, while European governments are busy with “immigration and poverty relief, supply-chain reform and other nonsense.”

3. Warsh wants to turn the clock back to 2008, but no one has successfully kicked the habit

• His reconstruction of Warsh’s intent: control only the short end, follow the inflation data, avoid QE, provide no 2-year guidance and do not adopt Japan-style YCC. Long-term rates would be left to the market, returning to the pre-2008 regime and using market pressure to force the government to “control the pace of its spending.” “That is a very idealized scenario.”
• Reality is different from 2008. The U.S. has a large volume of debt that must be issued continuously and is “at least at China’s 2013–14 position”—China has already “crossed to the other side of the bridge and burned it.” Once rate controls are loosened, long bonds will “enter a long bear market,” with rates grinding higher over time.
• The reason there may be no obvious reaction this year is that bank deregulation removes Treasuries from risk capital, allowing banks to lever up and buy them freely. That could create roughly $4.5T–$5T of additional deployable balance-sheet capacity. But the demand side needs a steady flow of new buyers every year. “If that is not solved, debt yields will keep rising.” The next few episodes will show whether Kevin actually follows through.

4. The final policy choice: defend the long end or inflation; the reservoir is capital markets

• YCC and QE are essentially money printing. To keep that printing from generating inflation, the Fed and European central banks spent years deleveraging the financial system. Banks had leverage ratios of 35 or even 40, which were brought down to 15; the deleveraging that began in 2008 was not finished until 2024–25. Now “there is not much room left for further deleveraging at the banking level.” The central bank therefore faces a choice: “To control long-end rates, you have to give up on inflation.” If it wants to control long-end rates without giving up on inflation, it must continue tightening financial regulation and push bank leverage even lower. “This is the Fed’s dilemma, and the final policy choice at this point.”
• Why has the White House cared so much about the stock market during the Iran episode and all along? “Because it has already staked everything on it.” Capital markets serve as the reservoir; China’s reservoir is real estate. The game can continue only if growth stays above rates. “But everyone knows the reservoir has a floor… debt will eventually blow up.”
• One camp within the GOP may be considering the opposite path: avoiding the old routes taken by China, Europe and Japan—“the 3 that went ahead did not do very well.” Japan combined extremely low growth, prolonged deflation and low rates with YCC to control rates, which is why its enormous debt has not yet caused a crisis; the cost was that “the entire economy lost its vitality.” If it refuses to take that old road, it cannot afford to give up on g, so “you place your hopes on AI” to deliver efficiency gains and sustain the capital-market reservoir. His conclusion was cautious: “We will see. I think the risks in the capital markets ahead are still considerable.”

5. July second-quarter earnings: CapEx and ROI must be “just right”; Google is the worst hit

• Cloud earnings will be judged on 2 things: CapEx guidance and ROI. “When the July earnings come around, there will be more and more discussion of ROI.” Google is the worst hit because it owns its model and operates in a fully internal loop; “both the front and back ends of the books will be laid bare.” The other providers’ revenue is coming from Anthropic and OpenAI. Neither model company is public, but “basically everyone knows the numbers,” so investors will have to assess whether AR growth is justifiable.
• The second ROI issue is cost review on the enterprise side. His own example: using an Anthropic model costs him more than $1,000 a day, and he uses it only to review code—he has not dared use it to write code. “AI intelligence belongs to the rich.” The best models go to the highest-gross-margin use cases; lower-margin businesses move into cost review. Tencent appears to have already reduced usage per employee. IT budgets cannot keep pace with model growth, and if a group of companies starts saying the same thing, ROI expectations will be affected.
• The CapEx dilemma: about $770B this year, with most valuations already priced through 2028 on a forward basis. “$1T next year, then $1.5T in 2028? Where does the money come from? Anthropic and OpenAI still need to raise money in the market, and how large is U.S. GDP?” If CapEx guidance is too high, the market crashes; if it is too low, “everyone immediately launches another round of semiconductor selling.” “You have to get it exactly right.” That is why he sees the major cloud providers’ second-quarter earnings as carrying “more downside than upside.”
• He raised his longer-term doubts about Google last year: “TPU is cheap, but performance per watt is far worse than NVIDIA’s cards. Why should its model remain ahead forever? I never thought its model could stay ahead forever.” “Google’s model still isn’t very good this year.” The broader issue is that these giants have harvested the world’s wealth for more than a decade and are now putting it into AI; perhaps 30% of that spending will flow outside the U.S. to Korea and Taiwan. That assumption is “extremely difficult to underwrite”—good luck.

6. Semiconductors are sold out across the board, individual names are fine—the risk is higher up

• Semiconductors themselves have “no problem whatsoever.” Every node from advanced to mature is in shortage; “if a company has no demand right now, it should simply die.” GPUs, CPUs, memory, optical modules and MLCCs are all sold out and all beating estimates. Cerebras is a new technology that OpenAI has taken separately; it comes down to how much supply you can secure.
• The market’s expectation for CapEx adjustment has no elasticity. Any sign of trouble—a CPU delay, slower 800V rollout or postponed deployment—will trigger a sharp drop. That is the defining feature of the first wave of a bubble: every decline attracts dip buyers betting on 2028 earnings and shortages that run all the way through. The conclusion is clear: “Do not focus on individual names; the individual names are fine.” Focus on cloud-company earnings. The real risk is higher up.

Verification Notes

• The original subtitles’ “new quality productive forces” and the wording that OpenAI “took Cerebras separately” are unclear in meaning; the literal phrasing has been retained.