Market Overview — September 23, 2025

Summary

• Although the Fed cut rates by 25 basis points, the speaker’s core call is that the terminal rate will be hard to push below 3%, because PCE remains around 3% and retail sales, corporate earnings, and household balance sheets have not deteriorated in tandem. White House fiscal, tariff, and immigration policies could either lift inflation or weaken growth, so the central bank needs to keep its options in reserve; he previously expected 2–3 cuts this year.
• He sees 4% as the floor for the 10-year Treasury yield, with the trade being to short bonds or buy swaptions. After the cut, 4.14% was “not particularly safe”; he views 4.10%–4.25% as a relatively safe, stable range at least through year-end. Heavy issuance, fiscal spending, and the lack of certainty that monetary policy will unleash large-scale liquidity together point to higher yields.
• The H-1B fee controversy is a snapshot of policy unpredictability and could also weaken America’s ability to absorb top international students. The Commerce Secretary initially said existing visas would also require a $100K annual fee and that transfers would be charged as well; two days later, the White House said existing holders would be exempt. If new hires still require an additional $100K, companies could cut a role that would have supported 3 employees to 2, making it “almost impossible” or extremely difficult for international graduates to stay in the US after graduation.
• China’s fiscal expansion remains a long-term theme, but the PBOC, CSRC, and relevant officials may not dare keep pushing asset prices unconditionally. The speaker says the stock market is “completely detached from fundamentals,” citing roughly RMB20T in local-government private financing, high leverage, and a scale contraction from roughly RMB450T to RMB350T; he also says “RMB350T of real estate” is providing support, though the original remarks did not clarify what the first set of figures referred to. Officials now need to “write a memorial to the emperor and report that all is well.”
• The TikTok deal and 2 explicitly mentioned $100B-scale arrangements are stitching Oracle, Nvidia, and OpenAI into a compute alliance. In the speaker’s reading, Oracle originally provided TikTok with cloud services and data storage; if the deal closes, the recommendation algorithm would also go to the US buyers. Oracle would buy roughly $100B of Nvidia compute, Nvidia would invest roughly $100B in OpenAI, and OpenAI would then purchase Oracle compute. This is not simply “the left foot stepping on the right foot”; it converts capital into resources for the AI compute war.
• AI competition is shifting from model quality to “total war,” where chips, power, data centers, and capital determine who stays at the table. A 10-gigawatt compute center uses roughly half the power of the Three Gorges project; aggregate capex at US tech companies has reached the trillion-dollar scale. The speaker favors Intel, which Nvidia has invested in and considers “severely undervalued,” arguing that it could become a supply alternative to TSMC and also force TSMC to expand capacity.
• ETH’s collapse alongside a gold rally leads the speaker to conclude that the move was not driven by monetary policy or insufficient liquidity, but looked more like an arbitrage hedge around a financing trade. Using BitMine’s roughly $300M financing as an example, he recalls that the stock was issued at roughly a 14% premium, around $70, with options attached; selling the stock and options at Friday’s prices could generate roughly 25%–30%. Subscribers could short ETH first, sell the stock after the open, and then cover the short, showing how mainstream crypto price action is increasingly shaped by Wall Street arbitrage.

Deep dive

1. Fiscal Expansion Keeps the Fed from Cutting Mechanically on Jobs Data

• The FOMC’s 25-basis-point cut matched the speaker’s prior expectation; he still sees 2–3 cuts for the year. Bostic’s hawkish view is that there may be only 1 more cut, and markets should not expect further easing next year.

• The central bank’s real dilemma is that fiscal stimulus has altered the traditional transmission mechanism: government spending and tax cuts improve corporate earnings, which lift equities and household assets such as 401(k) accounts. A weakening labor market therefore no longer necessarily brings consumption and inflation down in tandem.

• Unemployment could remain around 4.3%, while retail sales are “not that bad.” Immigration brings both consumption and labor; immigration policy could weaken nonfarm payrolls, while additional H-1B fees could suppress labor demand. Tariff and immigration policies therefore create two-sided risks: inflation could rise, or growth could fall short.

• PCE at roughly 3% or slightly below makes it difficult for the terminal rate to fall beneath 3%; without pushing PCE below 3%, deep rate cuts will be hard to deliver. Rates below 3% could also turn real interest rates negative, forcing renewed hikes if inflation reaccelerates. The speaker’s pointed conclusion: no Fed chair wants to go down in history for that.

2. White House Policy Reversals Force the Fed to Keep Its Options Open—and Add Long-End Supply Pressure

• The H-1B announcement captures the information environment facing the Fed. Lutnick initially said every existing H-1B visa would require a $100K fee, charged annually; transfers from Meta to Google, for example, would also be subject to the fee. Roughly 2 days later, White House spokesperson Leavitt explained that existing holders would not be charged. “It is not clear where this ends up.”

• If companies must pay another $100K for master’s or PhD graduates earning roughly $100K a year or less, the speaker’s projection is that employers may be forced to rely on US domestic labor: a company that could previously hire 3 people might hire only 2. The macro impact may be limited, but the policy could hurt America’s tech priority and weaken nonfarm payrolls.

• He believes 杰斐逊 and the next Fed chair, who takes office in 2026, will face the same problem: keeping options open while White House policy remains unclear. A 50-basis-point cut was impossible this time; retaining the option of another 25 basis points and watching how policy develops was the more rational approach.

3. Ten-Year Treasuries Are a Fiscal-and-Issuance Story, Not a Rate-Cut Story

• The speaker estimates that the “Big Beautiful Bill” could boost next year’s GDP by roughly 0.5 percentage points. Economic data may weaken in the fourth and first quarters, but government expenditure and AI capex mean the full year may not be poor.

• Growth resilience is arriving alongside heavy bond supply. With no certainty that monetary policy will unleash massive liquidity while fiscal spending continues, he viewed a 10-year yield around 4% as an extreme and recommended shorting bonds or buying swaptions: “4% is absolutely a limit… it will go higher.”

• After the yield rose to 4.14%, he said the level was “not particularly safe.” A range of 4.10%–4.25% is more likely to remain relatively safe and stable at least through year-end than to mark an immediate return to a sharp downtrend.

4. China Can Keep Fiscal Stimulus Going, but May Not Dare Use Money to Push Stocks Higher

• Looking back at last September 24’s “big package” and the stabilization-fund comments, the speaker acknowledged that stocks performed well over the past year but called it “a last gasp.” CPI remains negative, while the market continues to expect the central bank to print money and push up asset prices.

• His objection is not simply valuation. Local-government private financing has reached roughly RMB20T, while tier-2 and tier-3 cities still rely on financing to roll over projects. China’s deleveraging is incomplete, and leverage will be difficult to raise further. On the change in scale, he cited a figure behind total social financing falling from roughly RMB450T to RMB350T, and said “RMB350T of real estate” is providing support, without clarifying what the first figure specifically referred to.

• Unlike US companies, whose EPS and margins are both rising in the AI boom, China’s stock market is already “completely detached from fundamentals.” More liquidity cannot genuinely repair household and local-government balance sheets and could leave officials exposed to future accountability. The recent emphasis on value investing and long-term investing also signals internal concern.

• He summarizes the China-US difference this way: China’s central bank lacks independence and follows the Ministry of Finance on policy, while the US central bank does not know how the White House’s tariff, immigration, and fiscal policies will evolve. Both countries need a long fiscal cycle to keep growth above 2.5%, but both central banks have reached “a cautious point, an agonizing point.”

5. Oracle, Nvidia, and OpenAI Are Turning Capital Arrangements into a Compute Consortium

• In the speaker’s reading of the TikTok transaction, Oracle originally only provided cloud services and data storage; it could not use the data for large-scale training. If the deal closes as described, the board would apparently have 5 Americans and 2 Chinese members, formally retaining roughly 20%, while Oracle and Silver Lake would effectively acquire TikTok and transfer the recommendation algorithm to the US buyers.

• The subsequent funding chain is Oracle purchasing roughly $100B of Nvidia compute, Nvidia investing roughly $100B in OpenAI, and OpenAI then buying compute from Oracle and placing it on Oracle’s infrastructure. The objection is that this amounts to “the left foot stepping on the right foot.” His answer is that the money ultimately becomes deployable resources for the compute war rather than simply circulating in place.

• Sam Altman previously floated a roughly $7T vision for compute investment. The compute center established in the Nvidia-OpenAI transaction is 10 gigawatts, which the speaker roughly compares with half the power of the Three Gorges project. China’s recently discussed RMB500B financing, he says, appears to be part of total social financing, with one portion supporting small and midsize businesses and another supporting agriculture.

• The blocs he sees are Nvidia, Oracle, OpenAI, and OpenAI’s links with Apple on one side, versus Google and Broadcom’s TPU ecosystem on the other. Microsoft Copilot, Adobe, and others may still participate, but models and compute could ultimately be concentrated in a small number of conglomerates.

6. “Total War” Pushes AI Bottlenecks Toward Intel and Chip Capacity

• The speaker’s World War II analogy is that wars begin with elite troops but later become “total mobilization wars, where whoever has more capacity uses that capacity to crush the other side.” He uses the US construction of 166 aircraft carriers as a scale example. AI competition will likewise move from models and algorithms to who controls compute at scale.

• TSMC is the “oil field” in this analogy: it expands incrementally once customer demand is clear and will not take on enormous upfront risks based solely on an exponential-growth dream. According to the speaker, when Sam Altman previously sought massive capacity, 张忠谋 also believed such investment could not be driven by dreams alone.

• Nvidia’s investment in Intel could both force TSMC to reconsider its capacity plans and test a second chip supplier. The speaker has consistently viewed Intel as “severely undervalued” and said new information about Intel and other tech companies may emerge, while leaving the follow-up developments as an open question.

• Chip counts constrain data centers, and data centers in turn constrain model companies. As aggregate capex at US tech companies reaches the trillion-dollar scale, early participants such as Grok and xAI, which initially looked strong, could gradually fall behind for lack of the full stack, funding, and compute. The speaker’s conclusion remains explicitly subjective: “I personally feel” the endgame will be winner takes all.

7. ETH’s Crash Looks More Like Financing Arbitrage Than Monetary Policy or Liquidity Tightening

• Gold rising while crypto falls is the speaker’s evidence that the move was not caused by monetary policy or insufficient liquidity. He links the unusual price action to BitMine’s roughly $300M financing and believes Wall Street arbitrage and hedging may be an important factor.

• The financing, as he recalls it, issued stock at roughly a 14% premium, with the price around $70; the deal raised $300M, sold more than $300M of stock, and included options. Selling the stock and options at Friday’s closing price could generate roughly 25%–30%. “Give me access and I’m 100% in.”

• If the US stock market has not yet opened, arbitrageurs can short ETH to lock in the risk, then sell the stock after the open and cover the ETH short. Issuing common stock means the counterparty will directly sell most of its spot exposure, unlike Strategy’s model of raising money through bonds with attached options.

• This is also his broader view of how the crypto market is changing: long and short positions on Binance and macro narratives may not be the main sources of price discovery. The price action and price steering of many major coins “are already on Wall Street.” The market may have no genuine directional beta from ETH or from institutions buying large amounts of BNB; wherever an arbitrage opportunity exists, capital will take it.