The prospect of a September rate hike by the US Federal Reserve is not the decisive factor for long-term yields and equity market direction. In reality, recent price signals in the equity market suggest investors have not fully priced in tightening expectations.

High interest rates and rate hike expectations have a relatively minor impact on AI investments, while their potential negative effect on non-AI sectors is significantly greater. Theoretically, this should exacerbate the K-shaped market divergence, yet recent stock price movements tell a different story—particularly in the Chinese A-share market, where technology stocks have notably underperformed their non-tech counterparts. In an era of rapid AI technological advancement, the demand for government bonds—traditionally viewed as "safe assets"—should experience a structural decline. The selling of European and US government bonds is a consequence of economic and market operational logic, not a reason to predict short-term stock movements. Overall, we expect the market to remain range-bound, and investors should neither panic over overseas interest rates nor become overly aggressive simply because the market's reaction to rates is stronger than anticipated.

The widening of domestic and international long-term yield spreads is rooted in a deeper issue: a mismatch between capital supply and demand. As "commodity exports" encounter more potential frictions and disruptions, breaking the stalemate may require a pivot toward "financial exports."

Whether the Fed Raises Rates in September Does Not Determine the Direction of Long-Term Yields and Equity Markets

The recent rise in long-term US Treasury yields has been driven primarily by real interest rates, with inflation expectations contributing relatively little. This year, the 10-year nominal Treasury yield has climbed approximately 60 basis points, while the 10-year real yield has increased by around 50 basis points, accounting for over 83% of the total move. Moreover, the correlation between long-term Treasury yields and rate hike expectations has been unstable—it was weak from May to mid-July and again in late August, even turning negative at times. In other words, the rise in Treasury yields is not solely a function of Fed rate hike expectations.

The main driver of rising Treasury yields is the strengthening of financing demand from the US private sector, whose risk-adjusted returns now exceed those of public sector bonds, intensifying competition for capital between the two. As of July 2026, the trailing 12-month issuance of US corporate bonds reached $2.4 trillion, up 25.1% year-over-year, with its share of total debt issuance rising from 15.40% in early 2023 to 19.68%. Private sector financing demand continues to grow in weight within the bond market. We believe AI infrastructure expansion will further reinforce this trend. As long as cloud infrastructure returns for CSPs do not decline, debt financing instruments tied to AI infrastructure will crowd out demand for Treasuries, pushing long-term yields higher.

Consequently, either long-term yields continue to rise, potentially weighing on valuations in non-booming sectors, or the supply-demand balance in computing power eases, reducing cloud infrastructure returns and driving rates down while simultaneously lowering fundamental expectations—potentially the worse scenario. In either case, investing in equities is set to become considerably more challenging. Therefore, whether the Fed raises rates in September is not the core issue for long-term yields and equity markets; the most critical factor influencing their medium-term trajectory is the capacity expansion cycle of AI infrastructure.

High Rates and Hike Expectations Have Limited Impact on AI Investment, Yet Tech Stocks Are the Weakest Performers Recently

From a fundamental perspective, AI computing power investment is the least sensitive to Fed rate decisions and Treasury yields. North American CSP cloud business margins remain in a stable expansion channel; second-quarter 2026 earnings show operating margins for AWS, Google Cloud, and Intelligent Cloud at 39.4%, 35.6%, and 40.6%, respectively, continuing to rise quarter-over-quarter.

Large CSPs show low sensitivity to issuing debt at high rates. According to S&P and Moody's ratings, Microsoft is currently AAA/Aaa, Alphabet is AA+/Aa2, Amazon is AA/A1, and Meta is AA-/Aa3. Their interest-bearing debt/LTM EBITDA ratios range from 0.55 to 0.89 times, below the typical 1.0–1.5 times downgrade threshold used by rating agencies. Furthermore, per LSEG data, the five hyperscalers have issued approximately $223 billion in new bonds since 2026, exceeding the full-year 2025 total of around $109 billion and far surpassing the annual average of roughly $28 billion from 2020 to 2024.

Despite the significant increase in issuance, credit spreads on 2-4 year dollar bonds relative to Treasuries remain narrow. As of September 3, the median Z-spread for Alphabet, Amazon, and Meta's senior unsecured bonds with 2-4 years remaining maturity was just 34-36 bps (OAS median 51-53 bps), only slightly wider than the approximately 30 bps seen in 2025. This is more a technical consequence of increased supply rather than deteriorating credit quality. Moody's reiterated in its July report that the balance sheets of Microsoft, Alphabet, Amazon, and Meta rank among the strongest globally. This pattern aligns with every historical super-cycle driven by massive investment—whether China's 2006-2007 cycle or the US 2004-2006 property and financial cycle—where early rate hikes failed to dent demand. As long as computing power remains in short supply and the supply-demand gap persists, EBIT margins for computing infrastructure can be maintained, making the marginal impact of rising financing costs on income statements very limited.

From a market pricing perspective, the recent weakness in tech stocks reflects not current credit concerns at major tech firms but longer-term narrative and valuation issues. The question of whether "computing power advantages can translate into technological monopoly barriers" is the most influential factor in current tech stock pricing. It determines whether AI capital expenditure continues its "FOMO-driven expansion" or reverts to a traditional public infrastructure model. The core premise supporting the computing power investment narrative has been the advancement of model capabilities. Now, what's needed is both continued model improvement and a widening (and hard-to-narrow) gap between closed-source and open-source models.

We have previously noted that RSI (Recursive Self-Improvement) and anti-distillation could be two critical factors. If the story of AI training AI gains traction, it implies a substantial portion of incremental computing demand comes from AI agents themselves. If anti-distillation is convincing, it suggests latecomers will find it difficult to catch up with frontier models at low cost. This week brought two developments: OpenAI released GPT-6 Astra, suggesting RSI is initially being realized, and Anthropic released Claude Fable 5.1, which explicitly introduces anti-distillation mechanisms. We cannot yet determine whether these two model releases will immediately shift market narratives, as the effects of RSI or anti-distillation are not as readily observable by the market and public as Coding Agents or OpenClaw. There is currently a lack of direct and explicit metrics for measuring RSI and anti-distillation effectiveness. Both developments may help accelerate model iteration and widen model gaps, but whether they can drive an order-of-magnitude increase in computing demand, as agents did, remains uncertain. Until more evidence emerges, the market will continue to oscillate between the two long-term pricing frameworks—"computing infrastructure" versus "computing as a moat"—with interest rates serving as a short-term disturbance rather than the central issue.

Rate Hike Expectations Have a Greater Impact on Non-AI Sectors, Theoretically Exacerbating K-Shaped Divergence, But Recent Stock Performance Shows the Opposite

While rate hikes uniformly raise financing costs across the entire economy, sectoral differences in prosperity lead to varying price-demand curve slopes, with weaker non-AI industries suffering more significant demand damage. As Fed Chair Warsh noted at the Jackson Hole summit, more than half of this year's capital expenditure is contributed by AI-related industries, credit spreads have touched historical lows, and overall financial conditions are not restrictive. However, some rate-sensitive, non-AI sectors such as agriculture and real estate are already under pressure.

The equity market has followed similar logic, with strengthening rate hike expectations often intensifying the K-shaped divergence between AI and non-AI stocks. This was particularly evident in the second quarter, where the excess returns of core AI stock pools over core non-AI stock pools in China, the US, Japan, and South Korea moved largely in sync with market expectations for the Fed's December policy rate. However, in the two weeks surrounding the Jackson Hole summit, the K-shaped divergence in global markets did not continue to widen, despite rate hike expectations ticking up further during this period.

More notably, when the Chinese market opened on Monday, gold and non-ferrous metal sectors fell by a similar magnitude as US stocks did on Friday. Over the first two days of the week, their declines broadly tracked US equities, but Wednesday saw a significant rebound, generating excess returns relative to US stocks. Both Monday and Wednesday featured intraday "deep-V" reversals. The movement in precious metals sectors suggests equity investors do not genuinely believe the Fed will tighten persistently. The market appears to be betting either that the Fed won't raise rates in September or that stock prices have already priced in tightening expectations.

The Equity Market Has Not Aggressively Priced in Fed Rate Hike Expectations

The K-shaped divergence in global markets has not continued to intensify. Rate-sensitive Fed assets, such as non-ferrous metal stocks, have not experienced significant corrections. Rate futures imply weaker Fed hike expectations for this year than in May-June. Only long-term bond yields have hit new highs, while breakeven inflation expectations have slightly declined. This combination of asset price movements suggests the recent rise in long-term bond yields stems more from investors' reassessment of the intrinsic value of government bonds in Europe, the US, and Japan—driven by the crowding-out effect of AI investment on long-term government debt demand and a vote of no-confidence in fiscal policies—rather than from Fed rate policy itself.

We believe investors need not over-worry about the global bond sell-off; this may simply signal the beginning of an era of global capital scarcity and the end of the low-interest-rate era. In an age of rapid AI development, demand for government bonds as traditional "safe assets" should structurally decline. The selling of European and US government bonds is an outcome of economic and market operational logic and should not be used to predict stock market movements. Even if a short-term correlation exists, it is likely driven primarily by liquidity and sentiment effects.

As for potential short-term risks in the stock market, the main concern is that investors may not have fully priced in tightening expectations. The relatively strong performance of non-ferrous metal stocks suggests investors do not believe the Fed will raise rates. If a September hike does materialize and the market adjusts to absorb the potential sentiment shock, the impact of Fed rate policy and US long-term yields can subsequently be downplayed, allowing a return to fundamental factors. Conversely, if the Fed does not hike, the market may lack sufficient upside odds and certainty. Overall, we expect the market to remain range-bound—there is no need to panic over overseas rate issues, nor should one become overly aggressive merely because the market's reaction to rates is stronger than anticipated.

The Deeper Cause Behind Widening Domestic and International Yield Spreads Is Capital Supply-Demand Mismatch; Breaking the Stalemate Requires Shifting from "Commodity Exports" to "Financial Exports"

Except for China, virtually all major global economies are in a chronic state of insufficient savings. This condition was rational during the low-growth phase preceding the AI transformation, but the AI technological revolution has altered this dynamic. Even a dovish Fed cannot change this state of affairs. The variable that can genuinely break the deadlock lies in China's potential for "capital exports." China's excess savings flowing into global markets could alleviate the capital supply-demand gap created by AI investment, lower long-term rates, and support equity market valuations. Simultaneously, Chinese capital could capture higher expected returns, mitigating the "asset shortage" problem domestically.

Of course, for these "savings dividend" gains from domestic-foreign yield differentials to benefit the domestic economy, the government needs effective regulatory and tax collection measures to ensure overseas investment returns improve domestic fiscal revenues and enhance secondary distribution. This would ultimately influence consumption and alter the current state of low interest rates, low capital returns, and low inflation. This year, we have already seen preliminary signs: the state is strengthening overseas tax collection and tightening non-compliant, uncontrolled outbound investment channels. Viewing these measures purely as contractionary fiscal policies to boost tax revenue may be one-sided. We view them as preparatory steps to relax domestic capital outflows in a more compliant and monitorable form in the future, channeling a portion of the domestic-foreign capital return differential back to the domestic economy to improve fiscal positions, subsidize domestic demand, and create a new virtuous cycle. This is also the necessary path for China to enhance its long-term influence in global finance.

From this perspective, if the bull market of recent years was fundamentally driven by "commodity exports" (whether AI-related or not), then against an increasingly complex trade landscape, the next medium-term market rally will likely coincide with "Chinese capital exports" and "financial exports." The emergence of financial stocks—particularly non-bank financials—would be a crucial signal of this shift.

Risk Factors

Key risks include: intensified frictions between China and the US in technology, trade, and finance; domestic policy intensity, implementation effectiveness, or economic recovery falling short of expectations; overseas macro liquidity tightening more than expected; further escalation of conflicts in Russia-Ukraine and the Middle East; and delays in resolving China's real estate inventory overhang.