Lagarde tells MEPs Europe matches US on AI productivity gains despite investment gap

Christine Lagarde delivered a counter-intuitive message to the European Parliament's economy committee on Thursday: whatever Europe's deficit in building frontier artificial intelligence models, the continent is not falling behind the United States in the productivity gains that matter to firms and workers. The European Central Bank president told MEPs that the bank's own surveys and data show European companies, including small and medium-sized enterprises, are adopting existing AI tools at a pace that is generating productivity increases comparable to those recorded across the Atlantic.

The investment gap remains wide

The contrast with the funding landscape is stark. Figures published by the OECD in February show that of the $194 billion in global AI venture capital deployed in 2025, three-quarters went to US-based firms. The European Union attracted $15.8 billion, trailing China's $13.9 billion. That disparity reflects a structural reality: American capital dominates the expensive race to train ever-larger foundation models, while European investment has clustered around the application layer, embedding AI into established industrial, financial and service processes.

Lagarde acknowledged the divergence. "We're seeing significant investment in Europe as well for the diffusion of artificial intelligence in multiple sectors of the economy," she said, but she did not pretend the numbers were close. The ECB's interest lies less in who builds the models than in who uses them, and how quickly that usage translates into output per hour worked. On that measure, the early evidence suggests the diffusion phase is more evenly distributed than the invention phase.

Productivity data still in its infancy

The ECB president was careful to qualify her assessment. The data showing productivity gains is only "beginning" to emerge, she said, and the picture remains provisional. That caution is warranted. Macro-level productivity statistics are notoriously slow to capture technological shifts, the Solow paradox of the 1980s and 1990s, when computers appeared everywhere but in the productivity numbers, remains the standard reference point. Lagarde's claim rests on surveys and high-frequency indicators rather than national accounts revisions, which will take years to settle.

Still, the direction of the early signal matters. If European firms are indeed extracting similar gains from deployed AI as their US counterparts, the policy conversation shifts from "how do we build a European OpenAI?" to "how do we remove barriers to adoption across the single market?" The latter is a regulatory and skills challenge; the former is a capital and research challenge of a different order. Lagarde's testimony suggests the ECB sees the adoption story as the more immediate macroeconomic variable.

Labour market watch: no redundancy wave yet

The other side of the productivity coin is employment. Lagarde said the ECB would "look very carefully" for signs that AI adoption is displacing workers at scale. "So far, we are not yet seeing the waves of redundancies that are feared," she said, adding that the bank would be "extremely attentive going forward." That phrasing, "waves of redundancies", echoes a concern that has moved from academic literature into central bank speech in the past year.

The fear received a sharper articulation last week from Michael Barr, a governor of the US Federal Reserve. Speaking to a gathering of economists and analysts, Barr warned that rapid advances in generative AI could produce a "jobless boom" in which growth decouples from hiring, leaving many people "essentially unemployable." He urged policymakers to be "clear-eyed" about labour market risks as the technology expands. Lagarde did not dissent from the risk assessment but placed the ECB in observation mode rather than alarm mode. "We don't know yet what the impact on the labour market will be but we are looking very carefully," she said.

Monetary policy at the speed of silicon

The exchange with MEPs moved quickly from AI's microeconomic effects to its implications for the ECB's core mandate. Lagarde acknowledged that the velocity of technological change complicates monetary policy design. "The speed at which these technologies develop and update is incredible," she said. Whether the effects are "good or bad," the need for legislators and central bankers to adjust to this level of change is historically "new."

That admission is more significant than it sounds. Central banks have spent the past decade refining frameworks, forward guidance, strategic reviews, symmetric inflation targets, that assume a relatively stable structural backdrop. A general-purpose technology that rewrites production functions across sectors in quarters rather than decades breaks those assumptions. The transmission mechanism from policy rates to inflation may itself become less predictable if investment, productivity and labour supply all shift simultaneously.

Defending the data-dependent approach

Nikos Papandreou, a Greek centre-left MEP, pressed Lagarde on whether the ECB's cautious, "meeting-by-meeting" approach can still safeguard price stability amid new technologies, geopolitical turmoil and a worsening climate crisis. The question was a proxy for a broader debate: should central banks try to anticipate structural breaks, or wait until they appear in the data?

Lagarde was unmoved. "Some argue that we should stop being so reliant on data and focus more on anticipating developments we cannot be certain about," she said. But she added that she was "really convinced" that by staying data-dependent, the central bank remains "agile" and able to adapt to fast-changing risks as they unfold. The phrase "agile" is doing heavy lifting here. It implies that a framework built on realised data can pivot faster than one built on forecasts that may be wrong, a defensible position, but one that assumes the data arrives with sufficient lead time.

What the surveys actually capture

Lagarde referred to "our data and our surveys" without specifying which instruments. The ECB runs several relevant programmes: the Survey on the Access to Finance of Enterprises (SAFE), the Corporate Telephone Survey, and the newer Experimental Statistics on AI adoption. SAFE, conducted twice yearly, added AI-related questions in its 2024 waves. The Corporate Telephone Survey, a high-frequency panel of large euro-area firms, has included modules on digital investment and productivity expectations. Neither is a randomised controlled trial; both rely on self-reported expectations and perceptions. That is not a criticism, central banks have no other real-time lens, but it means the "productivity gains" Lagarde cites are, for now, anticipated or perceived gains rather than audited output-per-hour measurements.

The SME claim is particularly important. Small firms in Europe have historically lagged large ones in digital adoption, constrained by IT budgets, skills and the fixed costs of integration. If the ECB's data genuinely shows SMEs closing that gap with AI tools, many of which are now accessible via low-code platforms and API calls, it would represent a structural break. The next SAFE release, due in April, will be scrutinised for confirmation.

The transatlantic comparison problem

Comparing EU and US productivity dynamics in real time is fraught. The US Bureau of Labor Statistics publishes quarterly labour productivity data with a lag of roughly six weeks. Eurostat's equivalent comes later and is aggregated from national statistical offices with varying methodologies. The ECB's internal nowcasting models bridge the gap, but they are not public. When Lagarde says gains "appear similar across US and EU markets," she is almost certainly referencing the ECB's own nowcasts and the Federal Reserve's parallel work, not a harmonised dataset. The OECD's own productivity database, while authoritative, operates on an annual cycle and will not reflect 2025 developments until late 2026 at earliest.

This matters because the policy response differs if the similarity is real versus if it is a statistical mirage. If European firms really are extracting US-level gains from deployed AI, the priority is diffusion: standards, interoperability, data governance, skills. If the gains are overstated, if European adoption is broad but shallow, or concentrated in a few sectors, the productivity payoff may disappoint, and the ECB's optimistic signal could delay necessary structural reforms.