U.S. Bonds Face Sharp Sell-Off Amid AI-Driven Tech Rally

Editor’s Note: In September, the U.S. bond market experienced a sharp sell-off. The 10-year Treasury yield rose from 4.75% to 5.29%, increasing by 54 basis points in a single month, with some fixed-income assets declining by 2.3% to 5%. However, breaking down the yield movement reveals an easily overlooked fact: during the same period, the 10-year real rate increased by approximately 49 basis points, while the market-implied inflation compensation rose by only about 5 basis points. In other words, this bond market sell-off was not primarily driven by rising inflation expectations—what changed significantly was the real rate.

This raises a more complex question: Why is the market willing to demand higher real returns? Typically, rising real interest rates may be linked to stronger expectations of economic growth, adjustments in the future monetary policy path, or increased compensation for the risks of holding long-term bonds. More unusually, even as U.S. Treasuries plunged, AI-related tech stocks continued to perform strongly, with the semiconductor sector leading the gains, while most other stocks faced significant pressure.

Harry Mamaysky, founder of QuantStreet Capital, proposed in his latest monthly investment letter an explanation: the market may be repricing for stronger future economic growth and the capital demands arising from large tech companies’ ongoing expansion of AI infrastructure. Under this framework, higher interest rates and strong AI stocks are not necessarily contradictory, as investors may believe that future profit growth will be sufficient to offset the increased cost of financing and higher discount rates.

However, this remains an unverified market interpretation. Rising real rates do not necessarily imply improved growth prospects, and changes in the term premium suggest that investors may be demanding higher risk compensation. For equity and bond markets, the key is to distinguish whether this round of rate increases reflects higher future economic returns or simply makes holding long-term assets more expensive and risky.

The following is the translated text:

In September 2026, the U.S. financial markets exhibited a rather unusual trend: U.S. Treasuries experienced significant selling pressure, while technology stocks, particularly the semiconductor sector, continued to rise.

Main asset performance in September 2026: Technology and momentum strategies performed relatively strongly, while various bond categories and interest-rate-sensitive assets came under pressure.

According to U.S. Department of the Treasury data, the 10-year U.S. Treasury yield rose from 4.75% at the end of August to 5.29% at the end of September, increasing by 54 basis points in one month. QuantStreet Capital data shows that certain U.S. fixed-income assets declined by 2.3% to 5% during the same period.

However, the stock market did not experience synchronized, broad-based declines. Bitcoin and momentum strategies with significant holdings in semiconductors and technology companies performed strongly, while the Nasdaq Index remained elevated. Meanwhile, U.S. small- and mid-cap stocks, the equal-weight S&P 500 Index, and interest-rate-sensitive sectors such as real estate investment trusts, utilities, and financials faced selling pressure.

Typically, a significant rise in long-term interest rates increases corporate financing costs and reduces the present value of future profits, particularly harming high-valuation stocks. However, the current market divergence suggests investors may be pricing future returns for different assets in markedly different ways.

Harry Mamaysky, founder of QuantStreet, believes that to understand this market cycle, one must first answer a key question: What is the bond market really trading?

From the yield structure perspective, a key feature of the September U.S. Treasury sell-off was that nominal rates rose primarily due to higher real rates, rather than a simultaneous sharp increase in inflation compensation.

To understand this, it’s necessary to first distinguish between three concepts.

The nominal yield is the bond yield typically seen by investors, which includes compensation for expected inflation as well as requirements for real returns and other risk factors.

Real yield can be understood as the return after subtracting inflation compensation. In the U.S. Treasury market, the yield on Treasury Inflation-Protected Securities (TIPS) is commonly used to observe how the market prices real returns.

The difference between the two is the breakeven inflation rate, commonly used as a reference indicator for market-based long-term inflation expectations. However, it also incorporates factors such as inflation risk and liquidity, and is not equivalent to a pure inflation forecast.

The changes in U.S. Treasury yields in September can be further broken down. The nominal 10-year U.S. Treasury yield rose from 4.75% on August 31 to 5.29% on September 30, an increase of 54 basis points. Of this, the real yield increased from 2.44% to 2.93%, rising 49 basis points, while the implied inflation compensation rose only from 2.31% to 2.36%, an increase of 5 basis points.

September U.S. major bond ETF performance: Long-term Treasuries and certain credit bonds faced significant selling pressure.

This means that over 90% of the September rise in the 10-year U.S. Treasury yield was driven by higher real rates, rather than increased inflation compensation. At least based on this market indicator’s breakdown, the primary driver of the September bond market sell-off was not inflation compensation, but rather real rates.

This is particularly important because rising real interest rates and inflation compensation often correspond to different economic interpretations.

If the rise in yields is primarily driven by inflation compensation, it may indicate that investors are concerned about future price increases and a decline in the purchasing power of money, prompting them to demand higher nominal returns. However, if the increase is primarily due to real interest rates, further consideration must be given to expectations of economic growth, future real policy rates, and the risk premium investors require for holding long-term bonds.

This does not mean that inflation risks have disappeared. Inflation levels remain elevated, and oil prices, fiscal policy, and the Fed’s interest rate path continue to influence market expectations. However, based on the yield movements in September, attributing this U.S. Treasury sell-off solely to “worsening inflation concerns” is clearly insufficient.

Why has the real interest rate risen significantly? Mamaysky discusses several market explanations in his investment letter.

The first is that confidence in the U.S. dollar is being questioned. However, the dollar actually appreciated in September, which contradicts the narrative of a full-blown crisis of confidence in U.S. dollar assets.

The second is that investors have begun to worry about the U.S. government’s ability to repay its debts. The author argues that if market concerns about U.S. fiscal credit are primarily reflected through future inflation risks, long-term inflation compensation should show a more pronounced increase—but the September data does not exhibit this pattern.

However, this does not eliminate fiscal risk. An increase in government bond supply and higher risk compensation for holding bonds could also push up long-term yields, even if inflation compensation remains relatively stable.

In contrast, Mamaysky tends to favor another interpretation: the market may be pricing in stronger economic growth and reassessing the rising capital demands of large technology companies.

The key variable here is AI. As investment in AI infrastructure continues to expand, major cloud providers (hyperscalers) are pouring substantial funds into building data centers, procuring GPUs, expanding computing power, and configuring power and network infrastructure.

These expenses first imply a need for capital.

From a macro perspective, if companies aim to increase investment simultaneously while the supply of long-term funds available for allocation does not rise accordingly, upward pressure on funding costs may emerge. At the same time, if investors believe that AI will enhance future economic productivity and generate higher corporate profits, they may also raise their required rate of long-term real returns accordingly.

Both forces may be related to rising real interest rates, but their mechanisms are not identical: the former emphasizes capital demand and financing conditions, while the latter focuses on expectations of future economic returns.

Recent research by ING has also suggested a similar direction, indicating that AI's impact on bond yields stems not only from technology companies raising debt financing but also potentially from its reflection in real interest rates through productivity and long-term economic growth expectations.

However, such judgments still constitute market analysis rather than established causal relationships. An increase in real interest rates alone does not prove that AI is driving faster economic growth in the United States, nor does it confirm that AI-related funding demands are the primary factor behind U.S. Treasury sell-offs.

For Mamaysky, the appeal of this interpretation lies primarily in the stock market's reaction.

If U.S. Treasury yields rise sharply due to a deteriorating economic outlook, stock markets typically face broader pressure. However, in September, AI-related stocks such as semiconductors remained strong, indicating that investors still maintain optimism about the long-term growth of at least some technology companies.

This highlights a more interesting relationship between the bond and stock markets: higher real interest rates may be simultaneously priced in by the market alongside higher expected future earnings.

From a traditional valuation perspective, rising real long-term interest rates are typically not favorable for growth stocks.

Stock prices fundamentally depend on the discounted value of future cash flows. The higher the required rate of return demanded by the market, the lower the present value of the company’s future profits. This effect is typically more pronounced for growth-oriented companies whose profits are primarily concentrated in the future.

However, the performance of AI stocks in September suggests the market may be betting on another force.

In September, the semiconductor ETF (SMH) rose approximately 9.4%, while the equal-weight S&P 500 Index (SPW) fell about 4.8%, indicating a clear divergence between the technology sector and the broader market.

Mamaysky interprets this as a "numerator-versus-denominator" debate in valuation: rising discount rates increase the denominator, putting pressure on valuations; however, expectations of future profit growth increase the numerator, potentially offsetting some or all of these negative effects. In other words, the market may not be ignoring high interest rates—instead, it may believe that the future profit growth driven by AI will be sufficient to cover the higher cost of capital.

In September, momentum ETFs with significant holdings in companies such as AMD, Micron, Intel, Cisco, and Applied Materials performed strongly, with semiconductors continuing to be a key driver of market gains.

This trend has a solid fundamental rationale. AI infrastructure development first requires chips, servers, and related equipment, so upstream suppliers in the supply chain can secure orders and generate revenue earlier.

But this also raises the author’s biggest concern: while semiconductor stocks have continued to rise, the equal-weighted S&P 500 Index has underperformed. This highlights a clear divergence between investor sentiment toward the profit potential of AI infrastructure providers and expectations for the broader corporate sector.

Mamaysky refers to the broad corporate sector outside semiconductors as ROCS (Rest of the Corporate Sector).

In his view, companies that purchase AI chips are willing to invest large sums because they believe they will achieve economic returns through increased productivity in the future. The market is also willing to provide financing in advance for these yet-to-be-realized profits.

Therefore, it is not surprising that not all industries are growing in sync at this stage. What is truly puzzling is that the stock market itself is forward-looking. If investors are already confident that AI will significantly improve the future profitability of other companies, those expectations should gradually be reflected in the valuations of those companies.

However, the market in September did not experience this broad-based rally. While chip suppliers are already profiting, companies purchasing chips have not yet universally seen corresponding improvements in profitability. This means there is a business cycle in current AI trading that needs validation: the revenue gained by upstream companies must ultimately be supported by the ongoing economic value created by downstream enterprises. If AI fails to generate sufficient profits for a broader range of industries, rising expenditures on chip procurement, data center construction, and financing costs could gradually erode returns on investment.

Mamaysky does not believe that AI has formed a bubble. He still believes in AI’s long-term economic value, but thinks the market needs to see more evidence that AI’s benefits are spreading from the tech industry to other sectors.

The U.S. Bureau of Labor Statistics' productivity data has shown some positive signs, with recent productivity growth exceeding the long-term average since 2010. However, this improvement cannot be entirely attributed to AI, nor does it directly prove that businesses have generated sufficient new profits to cover their investment costs. From this perspective, both the bond and stock markets are essentially waiting for the same answer: Will future economic growth deliver the returns already priced in today?

Historical changes in labor productivity in the U.S. non-farm business sector.

Interpreting the rise in U.S. Treasury yields as a sign of increased market optimism about economic growth can indeed explain some asset price movements. However, this explanation still has a significant limitation: rising real rates do not necessarily equate to improved expectations for future economic growth.

The yield on long-term government bonds reflects not only investors' expectations of future short-term interest rates but also a term premium—the additional compensation investors demand for bearing risks such as price volatility associated with long-term bonds.

The term premium reflects interest rate uncertainty, fiscal supply, market supply and demand, and other risk factors. Even if inflation compensation does not rise significantly, long-term yields may still increase if investors are unwilling to lock in funds for extended periods and demand higher risk compensation.

This distinction is especially important in today's market.

A Reuters market analysis on October 7 noted that the term premium on U.S. 10-year Treasury bonds has risen to its highest level in about 12 years. This suggests that the rise in long-term yields may reflect not only expectations of economic growth but also investors’ reassessment of fiscal and monetary policy risks, as well as the risks associated with holding long-term bonds.

It is important to emphasize that real interest rates and term premiums are not two independent metrics that can be simply added together. The real yield on TIPS itself may already include a real term premium; therefore, the approximately 49-basis-point increase in real rates in September does not mean that all of this increase stems from stronger growth expectations.

Two different drivers can have distinct impacts on asset markets. If the rise in real interest rates primarily reflects improved expectations for economic growth, corporate future profits may rise in tandem, allowing some stocks to withstand higher discount rates.

However, if higher real interest rates and long-term yields are driven more by term premiums, companies may face persistently rising financing costs without a corresponding improvement in future profits. In this case, high interest rates would exert more direct pressure on stock valuations, bond prices, and corporate investment.

This is why we cannot conclude that the current U.S. Treasury sell-off is definitively a positive signal for economic growth simply because AI stocks are rising. For QuantStreet, the market currently lacks sufficient evidence to support a full shift toward any single asset class.

The institution remains relatively overweight in value stocks and low-volatility stocks, aiming to maintain exposure to a broader range of corporate sectors while continuing to hold certain technology stocks in portfolios with higher risk tolerance.

Bond allocations are also beginning to see minor adjustments. Mamaysky believes that the potential attractiveness of bonds has improved as the 10-year U.S. Treasury yield approaches around 5.25%. As a result, QuantStreet has begun to moderately extend duration in low-risk portfolios, increasing exposure to bonds that are more sensitive to interest rate changes.

However, this does not mean the institution has fully shifted to a bullish stance on long-term bonds. Its model still favors shorter-duration assets, and the overall bond duration remains below the benchmark, though the underweight position has narrowed.

The author also noted that for suitable investors, certain alternative assets, such as perpetual private equity funds, may offer some diversification benefits; however, the liquidity and valuation risks associated with these products must be considered separately.

These adjustments reflect a cautious stance: long-term yields have begun to appear somewhat attractive, but it remains unclear whether the forces driving yields higher have already subsided.

Next, the market needs to monitor three key signals: first, how long-term real interest rates and term premiums are evolving, to distinguish between growth expectations and risk compensation; second, whether AI investments are beginning to genuinely improve productivity, profit margins, and cash flows for non-tech companies; and third, whether Fed policy expectations, fiscal financing needs, and long-term Treasury supply continue to exert upward pressure on yields.

If economic growth and corporate earnings continue to improve, high real interest rates and strong stock performance may coexist for a period. However, if term premiums continue to rise and AI investment returns fail to materialize, tech stocks that currently appear resilient to high interest rates will face even greater valuation challenges.

The most important signal of the September U.S. Treasury sell-off was not that inflation expectations had spiraled out of control again, but that investors demanded significantly higher long-term real returns.

The real unresolved question is: Does this higher return requirement stem from confidence that the future economy will generate greater profits, or from increasing risks associated with holding long-term assets?

Both interpretations could push up U.S. Treasury yields, but they imply vastly different market outlooks for stocks, bonds, and the AI investment cycle.