A recent research report from CITIC SEC indicates that whether the Federal Reserve raises interest rates in September is not sufficient to determine the direction of long-end yields or the equity market. In fact, recent price signals in the equity market suggest that investors have not fully priced in rate hike expectations.
High interest rates and rate hike expectations have a relatively minor impact on AI investments, while the potential negative impact on non-AI sectors is greater. In theory, this would intensify the K-shaped divergence, but recent stock price movements show the opposite. This is particularly evident in the A-share market, where the technology sector has significantly underperformed non-technology sectors.
Overall, the A-share market is still expected to fluctuate within a range. Investors should not panic over overseas interest rate issues, nor should they become overly aggressive just because the market's reaction to rates has been stronger than anticipated.
Rate Hikes Not the Core Driver of Long-End Yields
The recent rise in long-end U.S. Treasury yields has been primarily driven by real interest rates, with the contribution from inflation expectations relatively limited. Since the start of the year, the 10-year nominal Treasury yield has risen by approximately 60 basis points, while the 10-year real yield has climbed about 50 basis points, accounting for over 83% of the total increase.
Additionally, the correlation between long-end Treasury yields and rate hike expectations has been unstable. From May to mid-July and again in mid-to-late August, the positive correlation weakened and even turned negative at times. In other words, the rise in Treasury yields is not solely attributable to Fed rate hike expectations.
The primary reason for the continuous climb in Treasury yields is the strengthening financing demand from the U.S. private sector, whose risk-adjusted returns now exceed those of public sector bonds. This has intensified competition for capital between the public and private sectors. As of July 2026, U.S. corporate bond issuance over the past 12 months reached $2.4 trillion, a 25.1% year-over-year increase, with its share of total debt issuance rising from 15.40% at the start of 2023 to 19.68%.
CITIC SEC believes that the expansion of AI infrastructure will further reinforce this trend. As long as cloud infrastructure returns for CSPs do not decline, AI-related debt financing instruments will continue to crowd out demand for Treasury bonds, pushing long-term yields higher.
Consequently, either long-end yields continue to rise, potentially suppressing valuations of non-booming sectors, or the supply-demand balance in computing power eases, reducing cloud infrastructure returns and lowering rates while simultaneously dampening fundamental expectations — which could be the worse scenario. In either case, the difficulty of equity market investing is set to increase significantly.
Therefore, the Federal Reserve's September decision is not the core contradiction determining the direction of long-end rates and the equity market. The most critical factor influencing the medium-term trajectory of both is the capacity deployment cycle of AI infrastructure.
Tech Sector Shows Weakest Recent Performance Despite Low Rate Sensitivity
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 reports show operating margins for AWS, Google Cloud, and Intelligent Cloud at 39.4%, 35.6%, and 40.6% respectively, continuing their sequential rise.
Large CSPs show very low sensitivity to issuing debt at high rates. According to S&P and Moody's rating data, Microsoft holds AAA/Aaa, Alphabet AA+/Aa2, Amazon AA/A1, and Meta AA-/Aa3. Their interest-bearing debt to 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.
Additionally, according to LSEG statistics, the five hyperscalers have issued approximately $223 billion in new bonds since 2026, surpassing the $109 billion issued in all of 2025 and far exceeding the annual average of about $28 billion from 2020 to 2024. Despite the significant increase in issuance, spreads on 2-4 year USD bonds relative to Treasuries remain narrow. As of September 3, the median Z-spread for Alphabet, Amazon, and Meta senior unsecured bonds with 2-4 years remaining maturity was only 34-36 bps (OAS median 51-53 bps), only slightly wider than the roughly 30 bps in 2025. This reflects technical pressure from increased supply rather than deteriorating credit quality. Moody's July report still emphasized that Microsoft, Alphabet, Amazon, and Meta possess some of the strongest balance sheets globally.
This pattern is consistent with every historical super-cycle driven by massive investment, whether China's 2006-2007 property-finance cycle or the U.S. cycle from 2004-2006. Early-stage rate hikes failed to damage demand in those periods. As long as computing power remains scarce and the supply-demand gap persists, EBIT margins for computing infrastructure can be maintained, and the marginal impact of higher financing costs on income statements remains very limited.
From a market pricing perspective, the recent weakness in tech stocks reflects not current credit concerns for major companies, but rather long-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 as "FOMO-driven expansion" or reverts to a traditional public infrastructure model.
The core premise previously supporting the computing power investment narrative was progress in model capabilities. What is now needed is both capability advancement and a widening gap between closed-source and open-source model performance that becomes difficult to close. CITIC SEC previously noted that RSI (Recursive Self-Improvement) and distillation prevention could be two important factors. If the story of AI training AI can be established, it would mean a significant portion of incremental computing demand comes from AI agents themselves. If distillation prevention can be demonstrated, it would mean latecomers would find it very difficult to catch up with frontier models at low cost.
This week saw two developments: OpenAI released GPT-6 Astra, beginning to tell the story that RSI is initially being realized, and Anthropic released Claude Fable 5.1, which explicitly introduces distillation prevention mechanisms. CITIC SEC cannot yet determine whether these two model products will immediately change the market narrative, as the effects of RSI or distillation prevention cannot be easily measured or perceived by the market and general public in the way that Coding Agent or OpenClaw can. Currently, there is a lack of direct and explicit metrics for RSI and distillation prevention effectiveness.
Both developments may help accelerate model iteration and widen the gap between models, but whether they can drive order-of-magnitude increases in computing demand like agents did remains uncertain. Until more evidence emerges, the market will continue to oscillate between two long-term pricing frameworks: "computing power as infrastructure" versus "computing power as barrier-building." Interest rates are merely a short-term disturbance in this process, not the main contradiction.
Rate Hike Expectations Impact Non-AI Sectors More, Yet Recent Market Action Shows the Opposite
Although rate hikes would theoretically raise financing costs across the entire economy equally, differences in industry prosperity lead to different price-demand curve slopes. Weaker non-AI industries would suffer greater demand damage. As Fed Chair Warsh noted at the Jackson Hole summit, AI-related industries contribute more than half of this year's capital expenditure. Credit spreads are at historical lows, and overall financial conditions are not restrictive, but some rate-sensitive, non-AI industries such as agriculture and real estate have already begun to show strain.
The equity market has followed similar logic, where strengthening rate hike expectations tend to intensify the K-shaped divergence between AI and non-AI stocks. This was particularly evident in the second quarter, when the excess returns of core AI stock pools relative to core non-AI stock pools across China, the U.S., Japan, and Korea moved 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 globally did not continue to widen, even though market expectations for Fed rate hikes warmed during this period. More notably, when the Chinese market opened on Monday of this week, gold and non-ferrous metals sectors fell by amounts comparable to the U.S. market's Friday decline. Over the first two days of the week, declines broadly matched U.S. markets, but Wednesday saw a clear rebound with outperformance relative to U.S. stocks. Both Monday and Wednesday featured intraday "deep V" reversals.
The movement in precious metals suggests stock investors do not genuinely believe the Fed will persist with tightening. The market either believes the Fed will not hike in September or that share prices have already priced in tightening expectations.
Equity Markets Are Not Aggressively Pricing Rate Hike Expectations
The K-shaped divergence has not continued to widen globally. Rate-sensitive assets such as non-ferrous metal stocks have not experienced significant corrections. Rate futures-implied expectations for Fed hikes this year are weaker than they were in May-June. Only long-end bond yields have hit new highs, while breakeven inflation expectations have declined slightly.
Looking at this combination of asset price movements, the rise in long-term bond yields appears more driven by investors' downward reassessment of government bond values in Europe, the U.S., and Japan — driven by AI investment crowding out long-term government debt demand and a vote of no confidence in fiscal policy — rather than by Fed rate policy itself.
CITIC SEC believes investors need not be overly anxious about the global government bond selloff. This may simply represent the beginning of an era of global capital scarcity and the end of the low-rate era. In the context of rapid AI technological development, the decline in demand for government bonds as traditional "safe assets" should be a secular trend. The selling of European and U.S. government bonds is a consequence of economic and market operational logic, not a reason to predict stock market movements. Even if short-term correlations exist, they are likely primarily related to liquidity and sentiment.
As for potential short-term risks in the stock market, the main concern is that investors may not have fully priced in tightening expectations. Given the recent strong performance of non-ferrous metal stocks, investors do not appear to believe the Fed will hike. On this basis, if a September hike materializes and the market reacts with a correction that absorbs the potential emotional impact, the influence of Fed rate policy and U.S. long-term yields can subsequently be downplayed, allowing a return to fundamental analysis. Conversely, if the Fed does not hike, the market may lack sufficient upside odds and certainty.
Overall, CITIC SEC maintains that the market will continue to fluctuate within a range. Investors should not panic over overseas interest rate issues, nor become overly aggressive simply because the market reacts more strongly to rates than expected.
Capital Supply-Demand Mismatch Behind Yield Spreads; "Financial Going Global" May Break the Deadlock
Almost all major global economies except China have been in a chronic state of insufficient savings. This state was reasonable during the low-growth phase before the AI transformation, but the AI technology revolution has changed this dynamic. Even if the Fed turns dovish, it cannot alter this condition.
The variable that can genuinely break this deadlock comes from China's "capital going global" potential. China's excess savings flowing into global markets could alleviate the capital supply-demand gap created by AI investment, reducing long-end yields and supporting equity market valuations. Simultaneously, Chinese capital could achieve higher expected returns, alleviating the "asset shortage" problem domestically.
Of course, for this "savings dividend" from domestic-foreign yield differentials to benefit the domestic economy, the government needs effective regulatory and tax collection measures so that overseas investment returns can improve domestic fiscal revenues and secondary distribution. This would eventually extend its impact to consumption, changing the current state of low rates, low capital returns, and low inflation.
This year, CITIC SEC has observed similar signs. The state is strengthening overseas tax collection enforcement while tightening non-compliant and uncontrolled overseas investment channels. Viewing these measures purely as contractionary fiscal policies to increase tax revenue may be one-sided. CITIC SEC believes these steps are designed to relax domestic capital outflows in a more compliant and monitorable form in the future, channeling part of the domestic-foreign capital return differential back to support domestic fiscal conditions and subsidize domestic demand, creating a new cycle. This is also an inevitable path for enhancing China's long-term influence in global finance.
From this perspective, if the fundamental basis of the past few years' bull market was "goods going global" — whether AI-related or not — then in the context of increasingly complex trade dynamics, the next medium-term market rally is highly likely to coincide with "Chinese capital going global" and "financial going global." The start of a rally in financial stocks, particularly non-bank financials, would be a critically important signal.
Comments