What exactly are global central banks concerned about as they collectively assess AI?

A technological revolution is transforming the way central banks observe the economy.

From capital input to industrial application and further to the financial ecosystem, AI (Artificial Intelligence) technology is iterating rapidly, bringing successive changes. However, the "sensory experience" of different individuals, enterprises, markets and even economies is not synchronized.

On October 19, 2026, Pan Gongsheng, Governor of the People's Bank of China, mentioned at the Annual Meeting of the Financial Street Forum that there may be a time lag between the demand and supply effects generated by artificial intelligence. In the short term, AI will improve micro efficiency, drive investment and push up inflation, while in the medium and long term, it can raise overall productivity and reduce inflation. He also pointed out that the artificial intelligence boom has become one of the main driving forces for the rise of global stock markets, the global financial market presents high correlation and high volatility, and proposed to focus on three key tasks in the next step: closely track and study the macro impact of the development of artificial intelligence, promote the financial support for the development of the artificial intelligence industry, and prudently advance the application of artificial intelligence in the financial sector.

Since the beginning of this year, the Federal Reserve, the European Central Bank, the Bank of Japan, the Bank of Canada, the Reserve Bank of Australia and the Bank of England have also assessed the impact of AI on the economy from different perspectives such as investment, prices, employment, interest rates and financial stability.

01

The "Puzzle" of Policy Response

Central banks around the world have different observation focuses. The Bank of Japan and the Reserve Bank of Australia have noticed the impact of AI demand on the prices of technology products; the European Central Bank pays attention to the gap between investment expansion, technology application and productivity; the Bank of Canada studies policy trade-offs under the changes in employment structure; the Federal Reserve discusses potential growth and labor, while the Bank of England incorporates AI financing into financial stability monitoring. All these studies are focused on issues that have not yet reached a unified conclusion, and do not mean that countries have adjusted their policies accordingly.

A phenomenon worth noting is that AI is leading to "multi-speed mismatch" in the economy (time difference, speed difference, expectation difference). With the continuous breakthrough of AI technical capabilities, capital expenditure has formed actual demand first, the financial market has priced in future returns in advance, while productivity improvement, employment adjustment and institutional adaptation are unfolding along different timetables.

This includes not only the familiar time lag of technology diffusion in economics, but also the cognitive and decision-making pressure brought by the accelerated evolution of new technologies. Investment growth does not necessarily mean the synchronous improvement of potential supply capacity, and market pricing cannot replace real cash returns.

For central banks, the crux of the problem lies right here — they need to not only judge what is happening in the current economy, but also distinguish which changes belong to short-term demand, which may form lasting supply improvement, and how the unfulfilled future will affect today's financial conditions in advance.

We may start from the starting point of AI investment. From chips, servers to data centers, power grids and cooling systems, enterprises purchase equipment, organize construction, and arrange energy supply and financing. These expenditures quickly form orders and current demand, but whether the completed equipment can be fully utilized and how much new revenue it can generate still needs gradual verification.

On July 28, 2026, the Bank for International Settlements (BIS) published the report *AI and the Global Economy: Implications for Central Banks*, pointing out that AI is driving large-scale investment expansion that is increasingly dependent on debt financing, and affecting the global economy through trade, financial markets and wealth effects. Although the potential productivity gains of AI are considerable, there are still uncertainties, and their distribution among different countries and industries is uneven.

In its July *Outlook for Economic Activity and Prices*, the Bank of Japan has included AI-related demand in its official forecast — this demand supports Japan's economic growth and also pushes up the prices of goods such as semiconductors. Affected by the combined factors of energy prices, yen depreciation and AI-related demand, the Bank of Japan expects that the year-on-year increase in consumer price index in the second half of fiscal 2026 may be significantly higher than 2%.

This does not mean that AI has become the main source of inflation in Japan, but it shows that a technology that is highly expected to reduce long-term costs is having an impact on actual prices.

The Reserve Bank of Australia pointed out in its monetary policy statement on September 29 that AI-related demand has driven the rapid rise in global technology product prices.

Capital input in Europe has also reached a visible scale. On October 13, Philip Lane, Member of the Executive Board of the European Central Bank, disclosed that enterprises in the euro area plan to allocate about 10% of their 2026 investment to AI, and AI-related loans account for about a quarter of the recent growth in corporate credit.

Lane also previously pointed out that if households and enterprises incorporate future productivity improvements into their consumption and investment decisions in advance, AI may create inflationary pressure through the demand channel in the early stage of transformation.

Why would a technology that may reduce production costs in the future increase the cost of some economic activities in the early stage of its development?

In his 2002 book *Technological Revolutions and Financial Capital*, techno-economist Carlota Perez divided major technological changes into the "installation period" and the "deployment period": capital first promotes infrastructure construction and industrial expansion, and technology can release productivity gains more widely only after adjustments in enterprise organization, business models and institutions.

From railways, electricity to information technology, there has been a time lag of varying degrees between capital formation and technology popularization. Perez's theory provides a historical reference for observing today's AI, but it cannot be used to assert that it will necessarily repeat the previous financial cycle.

What central banks need to distinguish is: does the current growth mainly come from construction activities, or have the conditions for the economy to continuously improve its supply capacity been formed? This will affect their judgment on the nature and persistence of inflation.

02

When Will Productivity Gains Be Realized

If changes in orders, investment and prices can be observed relatively quickly, the improvement of productivity usually takes a much longer time.

The results of the European Central Bank survey disclosed by Lane on October 13 show that 64% of the surveyed euro area enterprises have been experimenting with or using AI to a moderate extent, while only 7% consider their AI application level as "significant"; 52% of the surveyed workers said they use AI at work, compared with 26% in 2024.

Users have reported that a lot of time has been saved, but the benefits are mainly concentrated in specific tasks such as programming. The European Central Bank has not yet observed a significant and identifiable impact of AI on overall productivity.

A tool has entered the work schedules of more and more people, but why has it not left a prominent mark in macroeconomic statistics?

In a paper published in March this year by Bank of Canada researchers Joshua Brault, Maryam Haghighi and Jing Yang, this phenomenon is called the "AI productivity paradox". Some micro empirical studies show that AI can increase the productivity of specific tasks by 15%-60%, but the relevant benefits have not been fully reflected in macro productivity statistics.

These figures come from different tasks and samples, and cannot be directly used to estimate the efficiency improvement of the entire economy.

Classic studies provide another layer of explanation for this. In 2021, economists Erik Brynjolfsson, Daniel Rock and Chad Syverson published a study on the "Productivity J-Curve", which shows that after general-purpose technologies are put into use, enterprises usually need to adjust their business processes, train employees, transform organizations, and invest a large amount of intangible assets that are difficult to fully measure.

These upfront investments do not necessarily bring immediate and visible output growth. As new technologies and supporting organizations gradually mature, productivity gains may be further manifested.

Therefore, the increase in technology usage and the insignificant growth of macro productivity can coexist. However, statistical lags do not guarantee that a sharp productivity jump will definitely occur in the future.

Technological progress will also change labor demand.

The "two-speed economy" model established by the same Bank of Canada research team in July this year divides the impact of AI into two categories: assisting workers to improve efficiency, and automating tasks for substitution. The study finds that the impact on different industries may vary significantly. If the central bank tries to stabilize employment in affected industries by expanding aggregate demand, it may simultaneously increase inflationary pressure in other industries.

This is a policy trade-off revealed by the theoretical model, not a realistic judgment by the Bank of Canada that AI unemployment has already occurred.

In August, Kevin Warsh, former Chair of the Federal Reserve, also raised a series of unanswered questions in his speech at the Jackson Hole Economic Symposium: When will AI bring sustained economy-wide productivity improvement? Are computing power and labor complementary or competitive? Who will ultimately get the returns on technology investment?

These issues are related to the potential economic growth, and also to how central banks assess employment, wages and inflation.

In the final analysis, the long-term economic value of AI still depends on practical applications. Only when the improvement of model capabilities is further transformed into enterprise process optimization, production efficiency enhancement, and new products and services, can it be more widely included in national economic statistics.

03

Diverse Answers for Interest Rates

While changes in productivity and employment are still unfolding, the financial market is pricing for the future every day.

The *Financial Stability Report* released by the Bank of England in July points out that in 2025, the investment demand of some AI enterprises has exceeded the scale that their operating cash flow can support, and external financing further accelerated in the first half of 2026, involving multiple channels such as corporate bonds, bank loans, private credit and structured financing.

Market forecasts cited in the report show that the expected capital expenditure of large AI technology enterprises in 2028 has been revised up from less than 600 billion US dollars at the end of 2025 to more than 1 trillion US dollars.

The financing structure is changing the capital constraints of different enterprises in the AI industrial chain.

Data center buildings and supporting facilities can be used for a long time, but the internal chips and servers face faster technological iteration. Whether a long-term financing can be continuously repaid by the project's cash flow depends on equipment renewal, customer contracts, usage rate and future financing conditions.

Therefore, the Bank of England pays attention to how risks may be transmitted after AI construction shifts from internal corporate financing to the broad credit market. But the report also clearly points out that there is no sufficient evidence at present that AI financing has significantly crowded out the access to capital of other enterprises or governments.

In October, the *Global Financial Stability Report* released by the International Monetary Fund further warned that if AI investment and profit expectations fail to be fulfilled, asset price adjustments may amplify market volatility through leverage, concentrated positions and credit linkages.

This is a risk scenario that needs to be monitored, and does not mean that the technology investment cycle has reversed.

A more subtle change is taking place in interest rates. On October 13, Lane pointed out that the global AI investment expansion and the expectation of future productivity improvement may be exerting an upward impact on global long-term interest rates. The scale of AI investment in Europe is relatively limited, but it is also affected by the transmission of changes in the international bond market.

However, long-term interest rates are also affected by factors such as inflation, fiscal financing, savings, risk premium and monetary policy expectations. The increase in AI investment does not necessarily mean that the natural interest rate will rise.

If AI continuously improves productivity, it may expand profitable investment opportunities, as well as increase income and savings, and improve enterprises' debt repayment capacity. Its net impact on the long-term equilibrium real interest rate still needs to be tested by data.

Perez's distinction between financial capital and production capital is still enlightening in this regard: the capital market can price the technology prospects in advance, while enterprises must complete construction, sales and payment collection in a long production cycle.

Technology can continue to make breakthroughs, but asset prices do not necessarily rise synchronously. Changes in financing costs will in turn affect which projects can continue to be built and which innovations can get capital support.

This makes the relationship between AI and monetary policy more complex. Central banks need to not only observe the price pressure that investment demand may bring, but also assess the impact of technology diffusion on potential growth, and pay attention to new risks arising from credit expansion, financing structure and the use of AI by financial institutions themselves.

The policy direction put forward by Pan Gongsheng also has great practical significance. Pan Gongsheng pointed out that in the early stage, the People's Bank of China, together with relevant departments, studied the working ideas and directions for promoting the application of artificial intelligence in the financial sector from the aspects of improving infrastructure, enhancing application capabilities, expanding application scenarios and preventing and controlling security risks, and formulated and issued relevant policy documents on "artificial intelligence + finance". The People's Bank of China will continue to explore the application of artificial intelligence in areas such as assisting the analysis of monetary policy situations, financial risk identification, as well as payment, credit reporting, anti-money laundering regulatory services; at the same time, guide financial institutions to guard against risks such as data security and model application.

Major central banks around the world have begun to pay attention to, study and evaluate AI, and are facing several economic clocks running at different speeds: investment has taken place, technology applications are spreading, productivity gains are yet to be verified, while interest rates and asset prices are being formed every day.

Monetary policy needs to make judgments on current growth, prices and financial stability when the future has not yet been fulfilled.

This article is from the WeChat Official Account "Economic Observer", author: OUYANG Xiaohong, published with authorization from 36Kr.