CICC: "De-dollarization" remains an unstoppable trend

According to the Zhitong Finance APP, China International Capital Corporation (CICC) released a research report stating that "de-dollarization" remains a major trend: due to the continuous rise in U.S. government debt, which erodes the consensus on U.S. Treasury bonds as "safe assets," the short-term stabilization of the dollar's share in foreign reserves does not imply a change in the long-term trend. The decline in the status of reserve currencies often begins with a few countries reducing their holdings, which then gradually spreads to most countries, with the share of foreign reserves maintaining a downward trend amidst fluctuations. Stocks are information-sensitive assets, while government bonds are information-insensitive assets; the essential characteristics of these two types of assets are fundamentally different, making it difficult for U.S. stocks to replace U.S. Treasury bonds in fulfilling the function of reserve assets. The industrial trends of the AI revolution and the impact of debt have significant uncertainties, and there exists a "reverse Balassa-Samuelson" effect, which may not necessarily strengthen the dollar's status as an international currency.

CICC's main viewpoints are as follows:

The short-term rebound in the dollar's share of foreign reserves is not contradictory to the long-term downward trend.

By 2025, the dollar's share of globally allocated foreign exchange reserves is expected to decrease from 58.4% to 56.4%, but it is projected to rebound to 57.1% in the first quarter of 2026, seemingly reflecting a significant adjustment in global central banks' views on the dollar. However, after excluding the impact of exchange rate fluctuations of the dollar relative to other major currencies, the bank found that since the fourth quarter of 2024, the dollar's share of global central bank foreign exchange reserves has only experienced slight fluctuations and has not reversed the previously continuous downward trend, although there are indeed signs of a temporary slowdown. Analyzing the structure of global foreign reserves reveals that only a few countries are reducing their dollar reserves, while most countries have not significantly reduced their holdings. Some viewpoints argue that since the decline in the dollar's share of foreign reserves has stagnated and only a few countries are reducing their dollar holdings, "de-dollarization" is not valid (Goldberg and Hannaoui, 2024; Weiss, 2025).

The bank holds an opposing view, believing that the process of "de-dollarization" is still ongoing, and the temporary slowdown is not contradictory to the long-term trend. In the short term, only a few countries are reducing their dollar reserves, which aligns with historical patterns. During the decline of the pound's reserve status, the decrease in the pound's share of foreign reserves also began with adjustments from a few countries.

Before World War I, the pound's share of known official foreign exchange reserves fell from 64% in 1899 to 48%, primarily driven by a few countries such as Russia, Italy, and Greece increasing their holdings of francs, which correspondingly raised the franc's share of foreign reserves to 31% (Lindert, 1969). Although the pound's foreign reserve balance was surpassed by the dollar between 1924 and 1926, the largest holder of foreign reserves, France, accumulated a large amount of pounds between 1926 and 1927, which caused the pound's share of foreign reserves to rebound and exceed that of the dollar again. It was not until 1929 that France began to convert its pound reserves into gold, leading to another decline in the pound's share of foreign reserves. By 1931, other gold standard countries also began to sell pounds, and the reduction of the pound's holdings gradually spread from a few large holders to more countries (Eichengreen and Flandreau, 2009).

It can be seen that "de-poundization" continued for about 30 years from 1899, spreading from a few countries to most countries, during which the pound's share of foreign reserves also experienced periods of rebound. Therefore, merely relying on the small number of countries undergoing de-dollarization or observing a temporary stagnation in dollar reserve changes is insufficient to refute the long-term trend of de-dollarization In fact, "de-dollarization" and the decline in the proportion of dollar reserves reflect that the "anchor" of dollar hegemony is shaking, and the threshold for reversing this trend is high. The international reserve currency status of the dollar stems from the "dual anchors" of dollar hegemony: the "legal anchor" comes from the national credit of the United States, supported by economic strength and public institutions; the "functional anchor" comes from the vast, deep, and efficient dollar asset market. This supports two basic requirements for official foreign exchange reserves: the pursuit of credit quality, which requires reserve assets to maintain value over the long term and even appreciate during crises, manifested as "safe assets" (Brunnermeier et al., 2024); and the pursuit of liquidity, which requires assets to be readily convertible for external payments and market interventions. U.S. Treasury bonds possess both high credit quality and market liquidity, thus becoming the main asset for holding dollar foreign reserves.

However, due to the continuous expansion of U.S. Treasury bond issuance in recent years, the performance of dollar reserve assets has diverged: the dollar and short-term U.S. Treasury bonds still exhibit strong cash-like properties, meeting the needs for transactions, settlements, financing, and collateral, with relatively good performance; but long-term U.S. Treasury bonds have shown weaker performance, declining during crises such as the issuance of "reciprocal tariffs," the Greenland incident, and the U.S.-Iran conflict. The "weaponization of the dollar" has further increased the risks for foreign official institutions holding long-term U.S. Treasury bonds, driving them to turn to gold and other reserve assets (Arslanalp et al., 2023). Since 2020, the convenience yield of medium- and long-term U.S. Treasury bonds has significantly decreased and turned negative, while the convenience yield of the dollar itself has not contracted simultaneously (Du et al., 2025). This indicates that global investors still need dollar liquidity and are willing to pay a premium for it, but they no longer equally recognize the value retention capability of long-term U.S. Treasury bonds.

This divergence will shift the allocation of dollar reserve assets from a risk-hedging configuration to a liquidity configuration: although countries choose to maintain a large scale of dollar reserve assets, they simultaneously reduce the weight of long-term U.S. Treasury bonds in their reserves. Furthermore, the rebound in the share of dollar foreign reserves in early 2026 is also influenced by exchange rate valuations and changes in asset allocation, reflecting more short-term fluctuations and insufficient to prove that the reserve attractiveness of long-term U.S. Treasury bonds has been restored. The expansion of U.S. Treasury bonds and the shadow of dollar weaponization remain, so the long-term downward trend in the share of dollar foreign reserves will not change.

U.S. stocks are unlikely to replace U.S. Treasury bonds as reserve assets

In recent years, U.S. stocks have performed excellently, with the S&P 500 index continuously rising and reaching new highs, while risk premiums have continued to decline to nearly zero. Some viewpoints suggest that U.S. stocks could partially replace U.S. Treasury bonds, continuing to support the demand for dollar reserve assets: the United States has the largest, most liquid, and long-term high-return stock market in the world, allowing foreign official institutions to reduce their allocation to U.S. Treasury bonds and increase their allocation to U.S. stocks and corporate bonds, with U.S. corporate securities continuing to meet the demand for dollar reserves. From the data, foreign official investors have been continuously purchasing U.S. stocks, seemingly supporting this view. According to U.S. Treasury TIC data, from May 2025 to April 2026, foreign official institutions net bought $120.9 billion in U.S. stocks and $49.3 billion in corporate bonds, while net selling $17.5 billion in long-term U.S. Treasury bonds and $33.3 billion in agency bonds The bank believes that U.S. stocks, as risk assets, are difficult to replace the role of U.S. Treasury bonds as safe assets in global foreign exchange reserves. Holmström (2015) pointed out that stocks rely on continuous information production and price discovery, making them information-sensitive assets. Safe assets, on the other hand, need to maintain high credit quality and market liquidity, and preserve value during times of increased risk. This means that safe assets must be information-insensitive assets, exhibiting low volatility during economic and financial crises, so that official institutions can use them for external payments and exchange rate interventions at any time, fulfilling the function of reserve assets. Historical experience shows that during crises, the drawdown of stocks is significantly higher than that of government bonds. Baele et al. (2020) reviewed historical asset data from 23 countries and found that during financial crises, government bonds achieved an average excess return of 2.7% relative to stocks. After the trade policy shock in April 2025, U.S. stocks, U.S. Treasury bonds, and the U.S. dollar all fell simultaneously; U.S. stocks were sold off as risk appetite declined, failing to demonstrate the value-preserving ability during crises (Shin et al., 2025), making it difficult to replace them as safe assets.

Looking back at history, the experience of the yen illustrates that the currency demand attracted by high stock returns is difficult to traverse technological and asset price cycles. In the 1980s, Japan's electronics industry's competitiveness increased, and international capital inflows drove stock market valuations higher, leading to a continuous increase in the yen's share in global foreign reserves. In 1988, the price-to-earnings ratio of the Japanese stock market was about 32.6 times, exceeding the range explained by real interest rates and profit growth (French and Poterba, 1991). From 1990 onwards, the Japanese stock market experienced a long-term decline, and the yen's share of foreign reserves peaked at about 9% in 1991, subsequently declining continuously against the backdrop of Japan's long-term economic and financial stagnation, dropping to 3.9% by 2003 (Eichengreen and Mathieson, 2000; Park and Shin, 2012). It is evident that stock prices ultimately depend on technological prospects and corporate profits; once the return advantage disappears, international capital will be reallocated. Although risk assets can temporarily expand the international demand for a currency, they are difficult to sustain its status as a reserve currency in the long term.

Moreover, the allocation of U.S. stocks by some foreign official institutions is essentially still a pursuit of relative returns. In 2025, major global economies' stock markets generally outperformed U.S. stocks, and in the first half of 2026, the South Korean and Japanese stock markets temporarily outperformed U.S. stocks. This indicates that even in the context of the AI revolution, high-return assets will rotate among different markets according to profit cycles and industry trends, and will not remain concentrated in the U.S. stock market forever. For sovereign wealth funds, stabilization funds, and central bank income enhancement accounts, funds may flow in due to higher returns from U.S. stocks but will also be reallocated when relative returns improve in other markets. Such return-driven official capital inflows lack stability and are difficult to form the sustained dollar demand required for the core foreign reserves of central banks.

"AI Dollar" is unlikely to reverse "de-dollarization."

If the consensus on U.S. Treasury bonds as "safe assets" is shaken, and U.S. stocks cannot replace U.S. Treasury bonds in fulfilling official reserve functions, is there still another force that can reverse the trend of "de-dollarization"? Recently, a new viewpoint has emerged in the market, suggesting that the AI revolution will solidify the dollar's status as a reserve currency, primarily based on the logic that U.S. and allied enterprises control key nodes of advanced chips, cloud services, and computing power infrastructure Cross-border transactions can mainly be denominated in US dollars. Some viewpoints suggest that this could form an "AI dollar" similar to "petrodollars," increasing the demand for dollars and supporting dollar reserves (Ong, 2026). For the following reasons, the bank believes that the above conclusion is also difficult to establish:

First, the characteristics of computing power supply are significantly different from those of oil, making it difficult to form a long-term irreplaceability similar to oil. Oil is a natural resource that is universally needed by the global economy, and the distribution of production areas and transportation conditions cannot be changed in the short term, thus forming a natural monopoly in the oil production and sales process; however, models, chips, and computing power services will continuously have substitutes emerge with technological advancements and industrial competition, and the United States may not be able to maintain its lead and monopoly in the long term.

In 2025, private AI investment in the United States reached $285.9 billion, significantly higher than China's $12.4 billion, but the capital investment gap has not translated into a corresponding gap in model capabilities (Stanford HAI, 2026): In July 2026, the open-source large model Kimi K3 released by the Chinese AI large model company "Yue Zhi An Mian" rose to first place on the Arena code leaderboard and approached or exceeded the performance of leading closed-source models in several programming and agent evaluations (Arena, 2026; Moonshot AI, 2026). OpenRouter data shows that in 2025, US models contributed about three-quarters of the platform tokens; as of June 14, 2026, the weekly token share of Chinese models had risen to 55%, surpassing the 43% of US models (OpenRouter, 2026).

Second, the productivity gains brought by AI in the US may not support the dollar, but rather could lower the dollar through the "reverse Balassa-Samuelson" effect. Cerutti et al. (2025) argue that if productivity improvements from AI mainly occur in non-trade sectors such as education and healthcare, the prices in non-trade sectors will decline relative to those in trade sectors, leading to a decrease in the overall price level in the US relative to overseas, which would cause the real effective exchange rate of the dollar to decline. This is contrary to the traditional direction of the Balassa-Samuelson effect. Once the dollar enters a sustained depreciation channel, the share of foreign reserves will also be under pressure: exchange rate valuations will directly lower the proportion of dollar assets in global foreign reserves, and a weaker currency will also diminish the dollar's attractiveness as a store of value, prompting central banks to reduce their dollar allocations (Chinn and Frankel, 2007; Eichengreen et al., 2014). Therefore, AI does not inherently strengthen the dollar; if its result is a trend of dollar depreciation, it will instead accelerate the decline in the dollar's share of foreign reserves.

Finally, the impact of AI on the long-term debt repayment capacity of the US is highly uncertain. On one hand, if AI significantly improves total factor productivity, driving potential economic growth, corporate profits, and the tax base to continue rising, it would undoubtedly help improve the sustainability of US debt; on the other hand, the US debt burden itself continues to rise. The Congressional Budget Office (CBO) projects that the share of publicly held federal debt to GDP will rise from 101% in 2026 to 120% in 2036, while the federal budget deficit will expand from $1.9 trillion to $3.1 trillion during the same period. Therefore, whether the growth and fiscal revenue increments brought by AI can keep pace with or even reverse the trend of debt accumulation is currently difficult to determine At the same time, the impact of AI on employment structure and income distribution may require significant structural adjustments in fiscal policy. The government may need to increase taxes on AI-related capital gains, excess profits, and high-income groups, while expanding support for the unemployed, industries affected, and labor force transitions. If the new tax revenue can cover related expenditures, the AI transformation may help improve fiscal conditions; however, if taxation is insufficient and subsidies and social security expenditures grow rapidly, it may also expand the fiscal deficit over a certain period.

From this perspective, the ultimate impact of AI on the United States' debt repayment capacity remains uncertain, depending on multiple factors such as productivity improvements, technology diffusion, income distribution, and fiscal policy adjustments. Currently, it is insufficient to conclude that the sustainability of U.S. finances and the long-term value retention capability of the dollar will necessarily improve, nor can it be determined that the long-term downward trend in the dollar's share of foreign reserves will be reversed. Therefore, while the advantages of the AI industry can sustain the dollar's influence in transaction and investment phases, it is difficult to alleviate market concerns about dollar assets as foreign exchange reserves in the medium to long term.

In summary, the bank expects the long-term decline in the dollar's share of foreign reserves and the trend of "de-dollarization" to continue