Editor's Note: In September, a notable divergence emerged in the U.S. financial markets: Treasuries suffered a sharp selloff, with the 10-year Treasury yield rising more than 50 basis points in a single month, while some fixed-income assets fell 2.3% to 5%. At the same time, AI trades remained strong, semiconductor stocks continued to climb, and the Nasdaq held up relatively well, but small- and mid-cap stocks, the equal-weighted S&P 500 index, and other rate-sensitive sectors came under broad pressure.
Under traditional valuation logic, rising long-term rates mean future cash flows must be discounted at a higher rate, which usually weighs on equity valuations. Yet tech stocks did not correct to the same degree as the bond market. Is the market worried about the U.S. fiscal and inflation outlook, or is it betting that AI will deliver economic growth strong enough to offset the impact of higher rates?
Harry Mamaysky, founder of QuantStreet Capital, raised another question in his latest monthly investment letter: if the market believes AI can raise productivity across the entire economy, why are the main gainers still infrastructure suppliers such as semiconductors, rather than the broad set of companies that should eventually benefit from AI? The author does not believe AI has already formed a bubble, but remains wary of this divergence between earnings expectations and stock price performance.
This question bears on whether AI investment can form a sustainable commercial loop. Capital markets are already willing to pay high costs for computing power, chips, and data centers, but these inputs must ultimately translate into corporate profits to support long-term returns. As financing costs continue to rise, the AI investment thesis may increasingly depend on one question: who is bearing the construction costs, and who can truly capture the profits created by AI?
The following is a translation of the original text:
In September 2026, the most obvious change in the U.S. market was not the rise in AI stocks, but a rare divergence between bonds and stocks.
On one hand, the fixed-income market suffered broad selling. U.S. Treasury data show that the 10-year Treasury yield rose from 4.75% on August 31 to 5.29% on September 30, a cumulative increase of 54 basis points. According to QuantStreet's tally, some U.S. fixed-income assets fell 2.3% to 5% that month.
On the other hand, tech stocks, especially the semiconductor sector, remained strong. Bitcoin, momentum strategies with significant holdings in tech and semiconductor stocks, and the Nasdaq index all stood out relatively. By contrast, U.S. small- and mid-cap stocks, the equal-weighted S&P 500 index, and rate-sensitive sectors such as real estate investment trusts (REITs), utilities, and financials were broadly weak.
More unusually, the dollar and commodities rose simultaneously in September. This is inconsistent with the inverse relationship they typically exhibit, and suggests that the market environment investors face is becoming more complex.
For QuantStreet, there are two questions worth probing behind this round of market divergence: Why can tech stocks remain strong while long-term interest rates rise sharply? If investors are betting on economic growth driven by AI, why is this expectation not reflected in the broader stock market?
I. U.S. Treasury Yields Surge: What Exactly Is the Market Worried About?
The 10-year U.S. Treasury yield rose by more than 50 basis points in a month, meaning the pricing of long-term capital has undergone a notable change.
Bond prices and yields typically move inversely. Rising yields mean existing bond prices fall, and the longer the duration and the more sensitive the bond is to interest rate changes, the more pronounced the price pressure usually is.
But there is no explanation for what is driving this round of Treasury selling that has won consensus in the market.
The first explanation is that investors are losing confidence in the dollar.
However, Mamaysky is skeptical of this judgment. The dollar actually appreciated in September, which is inconsistent with the narrative of a broad sell-off in dollar assets. Of course, a rising dollar does not completely rule out long-term credit risk, but it at least shows that the market has not experienced a one-way crisis of confidence in the dollar.
The second explanation is that the U.S. fiscal position is prompting investors to demand higher long-term risk compensation.
The author likewise believes that the available evidence is not yet sufficient to support such a strong conclusion. He specifically notes that market-implied inflation breakevens have remained relatively stable, without changes of the same magnitude as the rise in yields.
This indicator reflects the yield differential between nominal Treasuries and inflation-protected Treasuries, and can be used to observe how investors are pricing future inflation and related risk compensation. If nominal yields rise while inflation compensation does not rise sharply in tandem, it is difficult to attribute the entire change to runaway inflation expectations.
That said, this does not mean fiscal risk can be ruled out. Long-term yields are also affected by factors such as Treasury supply, term premium, real interest rates, and market liquidity, and stable inflation compensation alone cannot prove that U.S. debt is free of risk.
By contrast, Mamaysky pays more attention to a third possibility: the market is repricing for stronger future economic growth and the enormous funding needs brought by AI infrastructure construction.
Large cloud service providers and tech companies continue to expand capital expenditure (Capex), building data centers, purchasing chips, and investing in supporting power infrastructure. These expenditures mean companies need to tie up more capital and may also increase financing needs.
If investors simultaneously expect AI to bring higher productivity and future earnings, then rising long-term interest rates may not necessarily be interpreted solely as a deterioration of economic risk, but could also partly reflect changes in growth expectations and capital demand.
This is not to say that AI investment has been proven to be the main cause of rising U.S. Treasury yields, but rather the author attempts to explain: why some tech stocks can still rise when the bond market is being sold off.
II. Why Are AI Stocks Stronger When Interest Rates Are Higher?
Stock valuation can be simply understood as the discounted value of future cash flows.
All else being equal, the higher the market interest rate, the higher the return investors demand, and the lower the value of a company's future earnings converted into today's dollars. This is also why high-valuation growth stocks are typically more sensitive to rising interest rates.
But the market in September did not fully operate according to this logic. Mamaysky's explanation is that investors may believe AI will create sufficiently strong future earnings growth to offset the valuation pressure from rising discount rates.
From the valuation formula perspective, this is equivalent to two forces competing with each other: one is rising discount rates, which压低 the present value of future earnings; the other is expected earnings increases, which raise the intrinsic value of stocks.
If the latter's growth is large enough, stock prices may continue to rise even in the face of higher interest rates. In other words, the market may not be ignoring high interest rates, but rather believes that future AI profits are sufficient to cover higher capital costs.
This explanation has some merit. AI infrastructure investment is creating enormous demand. Chip manufacturers, semiconductor equipment suppliers, and related tech companies can relatively directly generate revenue from construction spending. Companies such as AMD, Micron, Intel, Cisco, and Applied Materials also appear among the major holdings of the momentum ETF that the author focuses on. As long as the market believes AI capital expenditure will remain at high levels, earnings expectations for upstream suppliers may continue to be supported.
But the problem is that the revenue growth of these companies initially comes from other companies increasing capital expenditure, and does not necessarily mean the entire economy has already obtained corresponding productivity gains.
For companies buying chips and building data centers, the expenditure first forms costs or capital assets. Only when these assets ultimately help companies increase revenue, reduce costs, or improve profits can the investment generate sustained economic returns.
Therefore, the rise in semiconductor stocks only indicates that the market is optimistic about the profit prospects of companies related to AI infrastructure, and is not sufficient to prove that the ultimate economic returns of AI investment have already been realized.
III. The Biggest Question About AI: Chip Companies Are Making Money, What About Other Companies?
This is also the contradiction that Mamaysky focuses on most in his investment letter.
In September, the semiconductor industry performed strongly, but the equal-weighted S&P 500 index performed weakly.
Compared with the market-cap-weighted index, the equal-weighted S&P 500 index gives each constituent stock roughly the same weight, so it is better able to show whether market gains are broadly spread, rather than mainly driven by a handful of large technology companies.
The author refers to the broad group of companies outside semiconductors as ROCS (Rest of the Corporate Sector), that is, the rest of the corporate sector.
In his view, there is a logic in AI investment that requires time to verify.
Companies are now buying chips, servers, and software because they expect these technologies to bring productivity gains in the future. The market is willing to fund this buildout in advance also because it believes that unrealized profits will appear in the future.
Therefore, at this stage, it is not surprising that synchronized profit growth has not been seen across all industries.
The problem is that the stock market itself is forward-looking. If investors are convinced that AI will significantly improve the future profitability of other companies, then in theory, this part of the expectation should also gradually be reflected in the stock prices of the relevant companies.
But in September, there was no such broad-based rise. Semiconductor stocks continued to strengthen, while other companies did not receive similar valuation support. This led the author to ask: if the ultimate buyers of chips cannot obtain enough additional profit, how can they bear the increasingly massive AI procurement and buildout costs over the long term?
This question touches on the profit distribution mechanism in the AI investment chain.
In the short term, infrastructure suppliers may earn relatively high profits through order growth and tight demand. In the medium term, cloud service providers need to recover their investments by renting out computing power, providing AI services, and other means. In the longer term, ordinary companies need to turn AI into higher production efficiency, lower operating costs, or new sources of revenue.
Only when this process is gradually realized can the value created by AI investment potentially spread to broader economic activity.
Of course, the fact that stock prices have not risen in tandem does not mean that AI's productivity gains will not materialize. High interest rates, the industry's own operational pressures, and differing valuation starting points across companies may all be masking the market's expectations for future earnings improvement.
But at least judging from September's stock price performance, investors' confidence in AI upstream suppliers is clearly stronger than their pricing of the eventual benefits for other companies.
Regarding this phenomenon, Mamaysky has not concluded that AI investment is bound to fail. He still believes in AI's long-term value and does not think the current market already constitutes a bubble.
It is just that for this rally to gain broader fundamental support, it still needs to be seen that the profits created by AI are no longer confined to a handful of tech companies.
IV. What to Watch Next? Whether Productivity Gains Can Translate into Corporate Profits
In assessing AI's economic value, Mamaysky has begun paying closer attention to productivity data.
Revised data released by the U.S. Bureau of Labor Statistics (BLS) on September 3, 2026, showed that nonfarm business sector labor productivity grew at a quarter-over-quarter annualized rate of 1.4% in the second quarter, and 2.2% year-over-year.
From a longer-term perspective, from the fourth quarter of 2019 to the second quarter of 2026, U.S. nonfarm business sector labor productivity grew at an average annual rate of about 2.1%, higher than the roughly 1.5% level of the previous business cycle.
This is consistent with the productivity improvement trend observed by the author.
But it is important to distinguish that rising macroeconomic labor productivity does not equal confirmation that AI's contribution has been established. Capital investment, labor allocation, technological progress, and cyclical factors can all affect this indicator, and it is currently not possible to directly calculate from it how much profit AI has created for businesses.
Mamaysky therefore proposed a further verification standard: productivity improvements ultimately need to be reflected in the earnings of companies outside the tech sector.
This is also reflected in QuantStreet's portfolio adjustments. Although value stocks and low-volatility stocks performed poorly last quarter, the firm maintained a relative overweight in both categories, hoping to retain exposure to the broader corporate sector. At the same time, in portfolios with higher risk tolerance, it continued to retain some tech stock investments.
In the fixed-income market, the firm has also begun adjusting duration. Duration is used to measure the sensitivity of bond prices to changes in yields. The longer the duration, the greater the price loss typically faced when interest rates rise; but if yields fall, the potential price gains are also more pronounced.
The author believes that when the 10-year U.S. Treasury yield reaches approximately 5.25%, the potential investment appeal of bonds has begun to improve. Accordingly, QuantStreet slightly increased bond duration in its low-risk portfolio, marking a relatively notable directional adjustment for the institution in over a year.
However, this does not mean the institution has turned fully bullish on long-duration bonds. Its models remain unfavorable toward high-duration fixed-income assets, and the portfolio's overall duration remains below benchmark, though the degree of underweighting has narrowed somewhat.
For suitable investors, the author also mentions the diversification role of alternative assets such as Evergreen Private Equity Funds. According to the performance of some products cited, the relevant funds rose approximately 0.5% to 0.75% in September, providing a certain portfolio diversification effect during a month when most equities came under pressure. Nevertheless, such products still carry limitations including valuation frequency, liquidity, and underlying asset risks, and a single month's return cannot prove their long-term defensive capability.
These adjustments suggest that QuantStreet has not chosen to fully exit the AI trade, nor has it sharply pivoted to long-duration bonds simply because bond yields have risen. Instead, it is seeking a more balanced risk-reward across different assets.
What truly needs to be watched going forward are three sets of signals.
First, whether AI infrastructure spending can be sustained, and whether the revenue growth of upstream suppliers still has sufficiently strong demand support.
Second, whether AI is beginning to improve the profitability of non-technology companies. Productivity data can provide early clues, but corporate profit margins, cost savings, and incremental revenue are the more direct evidence of whether investment returns can be realized.
Finally, whether the rise in long-term U.S. Treasury yields reflects more of an expectation of economic growth, or rather inflation, fiscal supply, and term risk compensation. If growth fails to improve as expected while financing costs remain elevated, corporate investment returns will face greater pressure.
The question the current AI trade truly needs to answer is no longer just how much longer chip and computing power demand can grow, but how much incremental profit these investments can ultimately create for the entire economy.
Semiconductor companies have already generated visible revenue from capital expenditure, but broader corporate earnings improvement remains to be verified. Only when AI's productivity gains gradually translate into real profits outside the technology sector can the market obtain more complete evidence to support the current large-scale investment.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
