Dinner speech by Philip R. Lane, member of the ECB Executive Board, at the closing meeting of the European Central Banking System Research Network (ChaMP) on challenges to monetary policy communication in a changing world.[1]
Rome, July 6, 2026
First of all, I would like to congratulate everyone involved in the ChaMP research network on the remarkable success of the research program. This network has yielded many new insights into the transmission of monetary policy and has directly influenced policy debates in recent years.
In my remarks at this dinner, I would like to focus specifically on the topic of the impact of artificial intelligence (AI) on monetary policy stance.[2]
A natural benchmarking analysis is to see AI as permanently improving productivity and increasing income. The emergence and adoption of AI could put upward pressure on inflation through this demand mechanism already early in the transition phase, if households and businesses quickly recognize the permanent nature of productivity shocks and factor future income increases into their spending decisions.[3]
However, it is almost unrealistic to assume that households and firms know precisely the nature, magnitude, and persistence of future productivity shocks. A slower consumption response can also be rationalized if the level of delayed consumption is an important determinant of the benefits of current consumption, as in the “habit formation” model.[4] Consumers also face significant individual-specific uncertainty about the income impact of AI transitions, providing further reasons for delays in adjusting consumption.[5] It is more plausible to assume that households and firms will simultaneously learn the income and employment effects of productivity shocks over time and gradually adapt their spending accordingly.[6] In this case, the ex ante inflationary effect would be significantly reduced.
More generally, the inflationary effects of the AI transition will depend on a variety of factors, within the macroeconomic outcomes resulting from varying degrees of incorporating productivity and income gains into spending decisions.
One factor that will determine the impact on income, distribution, and demand is whether technological improvements from AI result in more labor or more capital. Technology is often modeled as labor augmentation, allowing more output to be produced with the same number of workers. This effect boosts workers’ labor incomes, but its magnitude depends on workers’ bargaining power and institutional factors. Conversely, if AI increases capital, the increase in income accrues to owners of capital rather than workers, thereby increasing inequality between labor and capital incomes.[7]Rising income and wealth inequality could limit the extent to which demand can grow in all sectors of the economy, weakening the inflationary trends associated with AI-driven productivity gains.[8]
The second factor is the scale of investment required to integrate AI into economic value chains. Significant computing power is likely to be required here, both to build the underlying models for AI and to implement AI in a business environment. Building the necessary computational infrastructure requires a significant increase in initial investment.
A third factor is that the expansion of AI-related computing will be accompanied by a significant increase in energy demand, putting upward pressure on energy prices until energy supplies catch up.[9] This dynamic could increase inflationary pressures during the AI adoption phase.
The geographic distribution of AI activities can be related to the impact on demand at a regional or national level. If AI activity remains concentrated in the US and China, and AI supply chains remain focused on Asia, European investment and energy demand growth will be relatively subdued. In this scenario, Europe would still face upward inflationary pressures from the effects of increased global demand for primary products and goods, particularly related to products used as inputs to AI production. In contrast, if technology diffusion into Europe is strong, these demand-promoting channels will function more strongly in the euro area. This is especially true where technology diffusion can only be achieved with some local capital investment.
Competing propositions about the expected macroeconomic impact of the AI transition can be translated into an impact on the natural rate of interest, defined as the real interest rate that matches desired savings and investment. On the one hand, continued optimism about AI-driven income and productivity gains will increase investment and reduce savings, putting upward pressure on R*. Conversely, the more uncertain households and businesses are about the trajectory of AI-driven income paths and the distribution of income gains across regions and income groups, the less likely it will be possible to increase R*. In particular, uncertainty around labor transfers and financing constraints for AI-related investments could increase precautionary savings.
The time profile of R* also depends on the trajectory of technology adoption. In one scenario, AI follows a typical S-shaped pattern, with adoption slowing in the early stages, accelerating as it reaches widespread adoption, and eventually plateauing as the technology matures. This profile means that AI permanently increases the level of productivity, but not the rate of productivity growth.
In contrast, an alternative scenario is that AI improves the innovation process, thereby moving the economy to permanently higher productivity growth rates. As long as productivity growth translates one-to-one into production and consumption growth, in the former scenario R* will eventually return to its pre-technological level as productivity growth weakens and the consumption growth path becomes lower again, whereas in the latter scenario R* will remain at a permanently high level.[10]
In either scenario, investment rates are expected to be highly volatile. One source of volatility is that demand complementarities may exist in the adoption of innovations, with each innovating sector benefiting if other sectors are also innovating.[11] Given the wide range of views on the long-term impact of AI, financial market sentiment towards AI-related investments may also be subject to waves of optimism and pessimism. In fact, multiple equilibria may exist, with a transition to a high-capital equilibrium self-validated by optimistic expectations that create a financing feedback loop.[12] In the transition to a high-capital equilibrium, investment increases rapidly and interest rates rise at first, but then interest rates fall sharply as capital becomes plentiful and income is generated primarily by capital owners who are more likely to save. At the same time, this mechanism is inherently fragile. A loss of confidence can lead to a self-fulfilling crash.
Finally, if AI production opportunities remain concentrated in the US and AI adoption rates are higher in China than in Europe, a scenario is possible where investors reallocate capital to both the US and China, reducing investment in Europe.[13] In particular, if foreign AI capital can still increase European productivity through licensing agreements, this scenario could still generate high revenues in Europe, but with relatively little domestic investment and downward pressure on Europe’s R*.
Several elements of this scenario are consistent with a high allocation to US technology stocks in euro area stock portfolios, high levels of imports of European intellectual property products from the US, and increasing substitutability of Chinese and European products in a wide range of middle-tech and high-tech sectors.
Given these different mechanisms, the ultimate impact of AI migration on R* remains uncertain.
So far, my discussion has focused on the impact of AI shocks on monetary policy. More broadly, it is also important to recognize the potential for the impact of AI to be amplified in conjunction with other cyclical shocks that can hit the economy. Let us outline three (perhaps interrelated) examples: (a) energy shocks; (b) Economic shocks. (c) Recession shock. The energy-intensive nature of AI means that continued upward energy price shocks could limit the rate of progress in building new AI models and reduce the rate of AI adoption. The capital intensity of AI production and AI deployment means that tighter financial conditions will also have a negative impact on AI production and AI usage sectors. Finally, AI could enhance workforce reductions during recessions by providing labor alternatives.[14]
It is clear that a feedback loop can exist between these three channels. For example, a sustained energy shock that changed the economics of AI production and deployment could also lead to a repricing of AI-related equity and debt in the financial system, a trend that would be further amplified if the economic downturn turns out to cause a larger-than-expected correction in the labor market, which also reduces consumption. A more resilient energy system will also reduce these risks, reinforcing the logic that the growing importance of energy-intensive AI sectors will accelerate the transition to renewable-based energy systems.[15]
In conclusion, in these remarks I have outlined the various channels through which AI may influence macroeconomic dynamics and monetary policy stances. Given the many uncertainties surrounding the strength and timing of different mechanisms, a data-driven approach is best suited to assess the overall impact of AI on the appropriate monetary policy stance. This will be a major challenge for financial economists and monetary policy makers in the coming years.
