The same confusion in China and the United States: can silicon-carbon shift from division to win-win?

The AI funding frenzy among tech giants is running into a cold shoulder from the bond market.

According to MarketAxess data, during this cycle, the prices of AI-related bonds with maturities of 10 years and above have continued to fall, becoming one of the worst-performing bonds in the investment-grade bond market.

A case that best reflects market sentiment comes from Amazon: on Tuesday this week, the company issued $25 billion in bonds, but demand for long-dated bonds was weak. The UK’s Financial Times, citing informed bankers and investors, said its five-year bond subscription orders were about 20% higher than those for its 30-year bonds.

In addition, the yield on a 30-year bond from SpaceX rose from the issuance price of 6.7% from less than two weeks ago to 7.3%.

According to Bank of America Global Research, for these five mega cloud service providers—Amazon, Google, Meta, Microsoft, and Oracle—their bond yields are currently about 0.6 percentage points higher than blue-chip bonds with the same credit rating and tenor. This risk premium is the highest across all industries in the investment-grade market.

Oversupply crushes demand

The immediate trigger for this round of sell-off is an unprecedented bond issuance wave by tech companies to fuel the AI arms race.

According to Bank of America Global Research statistics, this year to date, the cross-currency issuance scale of high-rated AI-related bonds has reached $270 billion, nearly double the full-year total from last year.

The ongoing influx of new bond supply has sharply increased pressure on investors’ holdings.

Janus Henderson’s global multi-asset credit head John Lloyd said that because many portfolios already hold large amounts of AI-related debt, investors have had to sell the mega cloud service provider bonds they already hold to free up room for Amazon’s new issuance. He said:

You have to offer enough concessions to attract us to participate in the new issuance.

The recent wild fluctuations in tech stocks have also suppressed market sentiment.

John Lloyd added that some investors already have substantial tech-sector exposure in their equity portfolios, which may further reduce their willingness to add related risk exposure in the bond market. Goldman Sachs Research Chief Credit Strategy Officer Amanda Lynam shared the same view.

Long-term returns in doubt, investors turn to the short end

Sell pressure has been concentrated in the long end, and the deeper logic is investors’ fundamental doubts about the long-term returns of AI capital expenditures.

DoubleLine portfolio manager Mariya Entina said:

Buying 30-year bonds usually requires companies to have very stable prospects and clear investment returns, and the long-term profit potential of AI capital expenditures is still in question.

She said the firm is more inclined to take risk that is more near-term.

Capital Group portfolio manager Pramod Atluri also prefers short-duration bonds from mega cloud service providers. Atluri said:

Technology iteration is so fast that long-term borrowing has become a higher-risk proposition. How the industry will look in ten years is something we simply cannot predict.

Mariya Entina also pointed out that the primary buyers of long-dated bonds are typically insurance companies and pension funds. These institutions need to match long-term liabilities, their investment style is often more conservative, and their tolerance for the above uncertainties is lower as well.

Higher-rate environment adds insult to injury

The appeal of mega cloud service provider long-dated bonds has also been further damaged by elevated short-end yields on U.S. Treasuries.

With inflation staying above the target level and market expectations that the Federal Reserve policy rate will remain “higher for longer”—especially after the hawkish signals released at the newly appointed Fed Chair Waller’s first meeting last month—short-term U.S. Treasuries have been offering quite attractive yields.

An analyst focused on high-grade credit said:

If you can lock in an attractive yield without going too far along the yield curve, why take on more risk?

At present, the short-term borrowing costs for mega cloud service providers remain stable, indicating the market is not worried about these companies’ near-term debt-servicing ability. But investors are making their stance clear with their actions: for whether this AI buildout wave can deliver its long-term promises, the bond market is far more cautious than the stock market.

Key takeaways: ① Stock market: extreme divergence between silicon-based and carbon-based. The U.S. stock market diverged earlier and China’s A-share market diverged even more; the root cause is that silicon-based earnings are clearly stronger than carbon-based. ② Real economy: the AI dividend has not yet benefited carbon-based businesses. AI capital expenditure has become a highlight for the U.S.-China economic picture; in China, Token call volume surged 81 times, yet the consumer confidence index fell to the historical 5th percentile. ③ Outlook: silicon and carbon will ultimately move toward a win-win. When AI moves from large models into physical applications, China’s smart manufacturing will have an engineering talent dividend and a compute-cost advantage, leading to a win-win integration of the new and old economies.

Same confusion in the U.S. and China: can silicon and carbon turn from divergence into a win-win?

Xun Yugen, Chief Economist at Guoxin Securities

Practicing qualification No.: S0980525090001

Tian Di, Macro Chief Analyst at Guoxin Securities

S0980524090003

Report release date: July 10, 2026

The most notable feature of the global economy and stock markets in the first half of this year is the divergence between silicon-based and carbon-based sectors. This article focuses on the U.S. and China—leaders of this AI technology revolution—analyzing the reasons behind silicon-carbon divergence and exploring pathways toward a future win-win.

1. Stock market performance: two extremes of “ice and fire”

Judging from the price gains and losses of silicon-based and carbon-based stocks, the U.S. stock market diverged earlier and China’s A-share market diverged more extremely. Using as samples the constituents of the CSI 300 and the S&P 500, grouped into silicon-based and carbon-based categories, the diverging trends are clear: silicon-based stocks in the U.S. started gaining momentum first. Beginning in early 2024, they continuously outperformed carbon-based stocks, and after 2025 the divergence intensified. In A shares, silicon-based stocks began to surpass carbon-based only in the second half of 2025, but then quickly widened the gap—especially in 26Q2, a “two extremes of ice and fire” period.

Note: In A shares, silicon-based companies include listed companies in CICC first-level industries such as communications, electronics, computers, power equipment, and new energy. Carbon-based companies include listed companies in 15 industries such as steel, building materials, construction, and transportation. Financials, machinery, and others are categorized as intermediate. In the U.S., silicon-based companies include listed companies in SIC second-level industries such as industrial and commercial machinery and computing equipment, measuring and analysis and control equipment, electronic equipment and components, communication services, and business services. Carbon-based companies include listed companies in 40 industries such as food and related products, general merchandise and retail, oil and natural gas production, and extraction. Financials and miscellaneous manufacturing are categorized as intermediate.

Industry price-change patterns also show that A-share divergence is greater than in the U.S. In the first half of 2026, among 30 A-share industries, half finished higher. The two major silicon-based industries—electronics and communications—were the leading drivers. Construction materials, which had the biggest gain, was mainly boosted by the electronic component supply-up driven by AI demand. Among the 15 declining industries, 8 carbon-based industries saw declines of more than 10%, and at the individual stock level, the proportion of declining stocks is as high as 71%. By contrast, among the 54 S&P 500 SIC second-level industries, 35 industries were up in the first half. Although the silicon-based industries did not rise as sharply as in A shares, they had stronger overall driving force on the broader market, pushing more than half of stocks across the U.S. market to close higher.

In terms of trading concentration, both A shares and the U.S. market have become highly aggregated toward silicon-based. Using the CSI 300 and S&P 500 constituent stocks to divide them into silicon-based and carbon-based for analysis, as of June 30, in A shares, silicon-based stocks accounted for 37.7% of the number of companies and contributed 63.1% of trading value; in the U.S., silicon-based stocks accounted for 33.3% of companies and reached 67.9% of trading value. Among them, the rise in heat for silicon-based stocks is more rapid in A shares. In the U.S., the share of silicon-based trading has remained above 50% for years since 2021; after 2023 it increased gradually. In A shares, before 2023 the proportion was still below 35%, starting in 2025 it rose noticeably, and in the first half of this year it quickly broke above 60%.

The stock price divergence between silicon-based and carbon-based sectors stems from divergence in earnings. From listed companies’ financial reports, A-share earnings divergence is obvious: in the silicon-based sector, all of them achieved positive year-over-year growth in attributable net profit for Q1, with half growing by more than 30%, while most carbon-based sectors lacked sufficient growth momentum. U.S. earnings distribution is more balanced: while silicon-based stocks led the gains, most carbon-based industries also showed strong performance. From industrial enterprise data, in China, from January to May, industrial enterprise profits in computer, communication, and other electronic equipment manufacturing grew year over year by 103.9%, far higher than the 18.8% for the entire industry. Meanwhile, profits in 20 carbon-based industries shrank by 15.4%. In contrast, in the U.S. in Q1, while silicon-based earnings led, the carbon-based sector performed relatively steadily.