The transformative energy of AI: Europe’s second to behave

Speech by Christine Lagarde, President of the ECB, BratislavAI Forum on synthetic intelligence and training as a part of an OECD high-level occasion to mark the twenty fifth anniversary of “Better Policies for Better Lives”, Bratislava

Bratislava, 24 November 2025

It’s a privilege to talk with you right this moment about synthetic intelligence.

In 1987 Robert Solow famously remarked that “you can see the computer age everywhere but in the productivity statistics.”

The identical remark might be made right this moment. We see AI advancing at exceptional pace. Yet its mixture influence remains to be barely seen within the knowledge.

Over the previous 12 months international company funding in AI reached USD 252 billion, and personal AI corporations raised a report USD 100 billion.^[1] Five main US buyers when it comes to capital expenditure at the moment are firms that focus closely on AI. None of those firms numbered among the many prime ten buyers a decade in the past. ^[2]

Some view this surge as non permanent exuberance operating forward of underlying fundamentals. But a debate framed solely when it comes to short-term ups and downs could miss the larger image.

History gives many examples of intense funding waves that – regardless of swings within the funding cycle – in the end left behind transformative applied sciences that reshaped economies for many years.^[3]

So the important thing query just isn’t whether or not there are cycles – that’s virtually sure – however how lengthy it is going to take earlier than the enduring productiveness advantages turn out to be seen.

And there are causes to imagine AI might unfold quicker, and ship tangible financial beneficial properties sooner, than earlier expertise waves.

If that’s the path we’re on – and I imagine it could be – Europe must place itself accordingly. We have to take away all of the obstacles that cease us from embracing this transformation. Otherwise we threat letting the wave of AI adoption cross us by and jeopardise Europe’s future.

History exhibits: disruptions first, advantages later

To perceive what’s at stake, it’s helpful to take a look at historical past.

Earlier common goal applied sciences, comparable to electrical energy, computer systems or the web, adopted a recognisable trajectory. Disruption arrived early, with broad-based productiveness beneficial properties solely rising slowly.^[4]

For instance, it took round thirty years earlier than the influence of electrical energy confirmed up clearly throughout the economic system. Power grids needed to be constructed, factories redesigned and employees reallocated from legacy duties to new ones.

Computers, too, required long-term investments in {hardware}, software program, abilities and new enterprise fashions earlier than they translated into measurable enhancements.

If Europe’s AI wave resembles the unfold of electrical energy within the Nineteen Twenties, annual productiveness progress might be about 1.3 share factors increased. But if it follows the US digital increase of the late Nineties, the increase could be nearer to 0.8 factors.^[5] Even that decrease certain could be important for Europe, marking a transparent step up from current development productiveness.

Could this time be completely different?

But AI has options that might compress this cycle and push ahead even larger productiveness beneficial properties. Two options – innovation and diffusion – level to a quicker path.

The first is that frontier innovation could speed up due to the recursive nature of AI.

AI methods can use their very own output to boost their efficiency in a steady loop. This can decrease not solely the price of producing items and companies, but in addition the price of producing new concepts.^[6]

For occasion, in fifty years, science resolved roughly 200,000 protein buildings. AI achieved over 200 million protein construction predictions in about one 12 months, vastly increasing the data frontier.^[7]

This represents a big change within the inputs to analysis and improvement. As the data base expands virtually in a single day, downstream discovery can compound sooner, even earlier than each lab or agency has totally reorganised.

By accelerating the manufacturing of concepts, AI can elevate not simply the degree of productiveness however doubtlessly the expansion price itself.^[8] Some estimates recommend that such AI-augmented R&D might double current US productiveness progress charges to between 1.6 and a pair of.4% yearly – quicker than earlier expertise waves.^[9]

Second, the diffusion of AI applied sciences might be quicker as a result of a lot of the supporting infrastructure already exists.

It is true that there are bottlenecks. The present wave of funding in hyperscalers exhibits that compute capability stays a constraint. Training and deploying bigger fashions requires substantial funding in knowledge centres and power. In Europe we face specific challenges on this respect, given our increased power prices and longer allowing delays.

But in contrast to previous applied sciences, comparable to electrical energy or computer systems that required new bodily networks or coding abilities, AI runs on present web gadgets and communicates with customers by human language.

Wide-scale use can due to this fact proceed even earlier than the infrastructure build-out is full. Many AI functions already ship beneficial properties on present {hardware}. So whereas a scarcity of computing capability holds again the tempo of mannequin improvement, it doesn’t essentially block diffusion throughout the broader economic system.^[10]

Moreover, the infrastructure itself is advancing rapidly. While Moore’s Law forecasts a doubling in chip capability each two years, AI mannequin compute energy has been doubling each six months – 4 instances quicker.

What Europe stands to realize

What does this imply for Europe?

The stakes might be terribly excessive.

With the United States and China forward of the sector, Europe has already missed the chance to be a primary mover in AI. And we nonetheless bear the prices of getting been gradual adopters over the last digital revolution. We can’t afford to make the identical mistake once more.

Yet the story is much from over. Europe can nonetheless emerge as a powerful second mover if it acts decisively. Our objective shouldn’t be to out-build the main AI fashions, however relatively to deploy AI throughout the board. By specializing in speedy adoption and good use of present AI applied sciences throughout our wide-ranging industries, Europe can flip a late begin right into a aggressive edge.^[11]

Our economic system is extremely diversified. The prime ten corporations within the US inventory market account for roughly 40% of the market throughout simply 4 sectors, whereas the highest ten within the EU account for not more than 18% throughout virtually twice as many sectors.

And European corporations are already adopting generative AI on the same scale to these within the United States. What the ECB is listening to from giant European firms confirms this development: many are investing closely in databases, cloud options and AI, with suppliers of those companies reporting double-digit progress.^[12]

But to show these advantages right into a aggressive benefit, we have to join knowledge throughout sectors. Thanks to industrial-scale knowledge areas, firms can share operational knowledge and create coaching units for AI fashions that no single agency might assemble alone.^[13]

Initiatives like Manufacturing-X and Catena-X within the automotive sector foster collaboration in knowledge sharing, whereas the European Health Data Space allows interoperable well being information, permitting us to leverage the broad anonymised affected person datasets generated by our common healthcare methods.^[14]

But these efforts won’t be sufficient on their very own.

If our knowledge areas use expertise stacks which are owned and ruled exterior Europe, we deepen – relatively than scale back – our strategic dependencies. We should diversify crucial elements of the AI provide chain and keep away from single factors of failure. In the foundational layers, comparable to compute capability based mostly on chips and knowledge centres, we must always preserve a minimal capability.

In the appliance layer, Europe ought to leverage the ability of the Single Market to implement interoperability and open requirements. This will encourage competitors amongst giant fashions and forestall the type of “lock-in” that has occurred with expertise platforms up to now.

Moreover, we should overcome a well-known set of outdated limitations which have prevented us from being first movers up to now.

If we enable our power prices to remain excessive, if laws stay fragmented, and if capital markets fail to combine and channel long-term, risk-bearing funding at scale, AI will diffuse extra slowly.

And this time, the results lengthen past dropping the race in AI fashions. We would ultimately face an extra lack of competitiveness for a lot of of our sectors and industries.

Conclusion

Let me conclude.

“It’ll be ten times bigger than the Industrial Revolution – and maybe ten times faster.” These phrases from Demis Hassabis – joint winner of the 2024 Nobel Prize in Chemistry for his AI analysis – seize the potential scale and pace of what could lie forward.

So the query is now not whether or not this new frontier will arrive, however how quickly – and the tempo of progress in recent times suggests it’s more likely to be before our establishments and laws are ready for.

That means performing now to clear the obstacles that might gradual AI diffusion and so delay prosperity for all Europeans within the many years forward.