若要在下一次工業革命中生存，歐盟應將 AI 變成公共品

Today, I want to write about something a little outside my usual topics.

As a European citizen, I worry about the possibility that the United States will achieve artificial general intelligence (AGI) before Europe does. I don't see the US as an enemy. Quite the opposite. But I live in the EU, and I don't want Europe to end up on the losing side of what could be the most important technological transformation in human history.

I think we're at risk of making a fundamental mistake in how we approach AI.

We're treating it as another technology sector in which European companies need to compete with American companies.

But what if AI becomes much more than that?

What if it becomes a foundational technology that transforms almost every industry, reshapes labour markets, changes how governments operate, and determines which countries accumulate wealth and influence over the coming decades?

We have seen something similar before.

The Industrial Revolution transformed the world. Countries that industrialised early gained enormous economic and strategic advantages. Others struggled to catch up, and some remained economically dependent on industrial powers for generations.

The historical picture is more complicated than a simple division between winners and losers. Some countries industrialised later and eventually converged with the leaders. But the broader lesson remains: when a technology transforms the foundations of economic activity, falling behind can have consequences that last for generations.

I believe AI could be another such turning point.

And if that is true, Europe should consider approaching AI very differently.

Instead of trying to compete with American AI companies solely by creating more startups, what if the EU treated foundational AI as a public good?

AI isn't just another technology sector

We are approaching a future in which AI could become as essential to economic activity as electricity, computing and internet access.

Today, we pay for software subscriptions because they provide useful tools. Tomorrow, AI could be embedded in almost everything we do.

Developers will use AI to write and maintain software. Doctors will use it to analyse medical information. Lawyers will use it to research cases. Engineers will use it to design products. Scientists will use it to accelerate research. Businesses will use AI agents to automate entire workflows.

Governments will use AI to deliver public services. Manufacturers will use it to optimise production. Financial institutions will use it to analyse markets and manage risk.

And if AI systems eventually become capable of performing a large share of economically valuable cognitive work, the transformation could extend far beyond what we currently imagine.

Entire business models could change. Some professions could shrink or disappear, while new ones emerge. The cost of starting and operating a company could fall dramatically. The balance between human labour, capital and intellectual property could shift.

Even people who never subscribe to an AI service directly will probably depend on AI-powered systems every day.

This is why I don't think AI should be viewed merely as a race to build the best chatbot or the most powerful model.

AI could become one of the foundations on which future economies operate.

And when a foundational technology becomes essential to economic participation, control over that technology becomes a question of national and economic power.

The lesson of the Industrial Revolution

Imagine a country during the Industrial Revolution that decides it doesn't need to develop its own industrial capabilities because it can simply import machinery from countries that have already industrialised.

Initially, that might seem perfectly reasonable.

Why spend enormous amounts of money developing factories, engineering expertise and industrial infrastructure when other countries already provide these things?

The country could import machinery, benefit from cheaper manufactured goods and enjoy some of the productivity improvements without bearing the full cost of industrialisation.

But over time, a problem could emerge.

The industrial leaders would accumulate capital, expertise, manufacturing capacity and the ability to improve their technologies. Their industries would become more productive, allowing them to reinvest in further improvements.

Meanwhile, the importing country might become dependent on external suppliers for essential capabilities.

It could still benefit from industrialisation. But it would have less control over the technologies that increasingly determined its economic future.

AI could create a similar dynamic, potentially on an even broader scale.

The countries and companies that develop the most capable AI systems could gain advantages across many different industries simultaneously.

They could use AI to develop better software, accelerate scientific discoveries, design new products, improve manufacturing and create more capable AI systems.

If these advantages compound over time, the gap between leaders and followers could widen.

Countries that depend entirely on foreign AI might find themselves in a difficult position: their economies could become more productive while a disproportionate share of the profits, intellectual property and strategic advantages flows elsewhere.

Of course, importing technology is not inherently bad. Countries can learn, adapt and catch up. Access to foreign technology can also accelerate development.

But there is a difference between using technology developed elsewhere and having no meaningful alternative when access to that technology becomes strategically important.

The danger isn't that Europe might lose the AI race. It's that Europe might lose the ability to shape the world that emerges from it.

What happens if AI becomes a utility?

Look at cloud computing.

A significant part of the infrastructure European businesses depend on is provided by American companies. This isn't necessarily a bad thing. These companies have invested enormous amounts of money in building reliable infrastructure, and European businesses benefit from it.

But it creates a dependency.

Now imagine that AI becomes much more important than traditional cloud computing.

Instead of renting servers to host an application, businesses will rent access to intelligence capable of performing a substantial part of the work their employees do today.

The economic value of that intelligence could be enormous.

If a company uses AI to generate millions of euros in additional value, a portion of that value will flow to whoever provides the AI. Scale this across millions of businesses and hundreds of millions of people, and the implications become difficult to ignore.

The problem isn't that American companies make money. They should be able to compete and profit from their innovations.

The problem is what happens if Europe becomes structurally dependent on foreign AI providers and never develops competitive alternatives of its own.

Imagine that European businesses become increasingly reliant on AI systems whose most important capabilities, infrastructure and strategic decisions are controlled elsewhere.

Imagine that European startups must compete against companies with privileged access to the most powerful AI systems, while paying for the same underlying capabilities through external providers.

Imagine that European governments depend on AI infrastructure they cannot fully control.

Europe might still be wealthy. It might still have excellent universities, talented engineers and successful companies.

But it could gradually lose the ability to determine the terms on which its economy accesses one of its most important productive resources.

This is not simply a question of technology spending. It is a question of economic independence.

What if AI becomes a public good?

Imagine the EU decided that foundational AI infrastructure was strategically important enough to justify substantial public investment.

The idea would be relatively simple.

The EU would invest in computing infrastructure, fund the development of competitive open-weight AI models, and provide European citizens and businesses with access to those models at little or no cost.

Think of it as public infrastructure for the AI economy.

We already accept that governments should invest in roads, electricity grids, scientific research and education because their benefits extend far beyond the organisations that directly use them.

These investments help entire economies function. They create opportunities for businesses, reduce barriers to participation and provide a foundation on which private enterprise can flourish.

Why shouldn't foundational AI be considered in a similar way?

Of course, AI is not exactly like a road or an electricity grid. Models require continuous research, expensive computing resources, maintenance, energy and specialised talent. Providing unlimited AI inference to everyone for free would be extremely expensive.

But the underlying principle still applies.

If access to AI becomes essential for economic participation, there is a strong argument for ensuring that everyone has access to a capable baseline of AI without having to pay a private subscription.

More importantly, Europe should have the infrastructure, expertise and technical capabilities required to maintain that access independently.

The goal would not be to make every AI product free. It would be to ensure that access to foundational intelligence does not depend entirely on a small number of private providers.

How would this work in practice?

I imagine four main components.

1. EU-owned AI infrastructure

The EU and its member states would invest in strategic computing infrastructure.

This would include data centres, accelerators, networking and the energy infrastructure required to operate them.

The goal would not be to replace every private cloud provider. It would be to guarantee that Europe has sufficient computing capacity to train, run and improve its own AI models.

The infrastructure could be operated by specialised organisations under public ownership and transparent governance, with private companies continuing to build and operate infrastructure alongside it.

Europe would not need to become completely self-sufficient in every component. That would be unrealistic. But it should avoid a situation in which it has no viable alternatives for capabilities essential to its economy.

2. Open-weight European AI models

The EU would fund research teams to develop competitive foundational models.

These models would be released with open weights and licences that allow researchers, startups and businesses to use and adapt them.

This would not mean that every model must be developed by a government department. Public funding could support universities, research institutes, startups and private companies.

The important thing is that Europe would build and retain meaningful control over its AI capabilities.

Open weights would also allow businesses to run models on their own infrastructure when appropriate, reducing dependence on a single provider.

The objective would not necessarily be to produce the most powerful model in the world. It would be to ensure that Europe has access to models capable enough to support its economy, and that it can continue improving those models over time.

3. Free AI access for European citizens, with reasonable limits

Every EU citizen should have access to a public AI service at no cost.

The service could help with programming, education, research, writing, translation, administration and many other everyday tasks. The objective would be to ensure that access to capable AI does not become a privilege reserved for people who can afford multiple subscriptions.

However, free access should not mean unlimited access.

Operating AI systems requires computing power, energy and ongoing maintenance. Without reasonable limits, a small number of users could consume a disproportionate share of public resources, while automated abuse could generate enormous costs for taxpayers.

The public service should therefore include fair usage limits, rate limits and safeguards against automated exploitation. Users who need additional capacity could pay for it or access specialised programmes where there is a clear public interest.

The limits should be generous enough to support meaningful everyday use, learning, research and personal projects, rather than making the free service so restricted that it becomes useless.

There should also be clear rules governing privacy, data retention and acceptable use. Citizens should know how their information is handled, and the service should not become a mechanism for unnecessary government surveillance.

The principle is straightforward: everyone should have access to a capable baseline of AI, but no individual should be able to exploit a publicly funded service at everyone else's expense.

4. Free AI for European businesses and EU employees, with accountability

I would go a step further for European businesses and people working in EU institutions.

Eligible European businesses should be able to access public AI models and computing resources without paying licensing fees or standard usage charges, subject to reasonable usage policies and capacity limits where necessary.

EU institutions and their employees should also have access to these capabilities for legitimate official work.

The reasoning is simple: if AI becomes a foundational technology, public investment should help European businesses adopt it, improve their productivity and develop new products and services. Publicly funded AI should also help European institutions operate more efficiently and deliver better services.

Imagine a small European startup that can build and test an AI-powered product without having to spend thousands of euros on proprietary APIs. Or a small manufacturer that can use AI to improve its operations without facing an additional technology cost it cannot afford.

This could lower barriers to innovation and make it easier for European businesses to compete internationally.

But there is an important condition: free access must not become an unchecked subsidy for private interests.

The infrastructure would be funded by taxpayers, so its use should be accountable to the public.

Businesses receiving free access should agree to transparent eligibility requirements and proportionate audit procedures. These could include usage records, checks on organisational eligibility and investigations into suspicious or excessive consumption.

The aim would be to identify and prevent abuse, such as reselling publicly funded compute without authorisation, creating unlimited accounts to bypass usage limits, using the service primarily for unrelated commercial exploitation, or diverting resources to ineligible entities.

Audits should be risk-based rather than unnecessarily burdensome. A small business using AI for ordinary development work should not face the same scrutiny as an organisation consuming enormous amounts of computing capacity or exhibiting suspicious usage patterns.

Sensitive business information must remain protected, and audits should focus on eligibility, resource consumption and compliance with the programme's rules rather than giving the government unrestricted access to private company data.

EU employees should similarly receive free access for legitimate work purposes, subject to institutional security policies, appropriate data handling and auditable usage controls. Access should not become a way to process classified, confidential or personal information through systems that have not been approved for that purpose.

There should also be clear consequences for deliberate abuse, including suspension of access, recovery of improperly obtained benefits where legally appropriate, and stronger penalties for serious or repeated violations.

The objective is not to burden businesses with bureaucracy. It is to ensure that public resources are used for their intended purpose.

If the EU invests in AI infrastructure to improve European productivity and technological independence, it should make that infrastructure genuinely useful to businesses while ensuring that the benefits are not captured through misuse by a small number of organisations.

Free access should be generous, practical and accountable.

Would the economic benefits outweigh the investment?

This is the most important question.

Building competitive AI infrastructure would require enormous investment. Training advanced models, purchasing accelerators, operating data centres and attracting researchers are all expensive.

But I think we need to look at the return differently.

The goal would not be to generate a financial return from selling AI subscriptions. The goal would be to increase the productivity and resilience of the entire European economy.

Suppose public AI infrastructure enables millions of workers to become more productive.

Businesses develop products faster. Researchers make discoveries sooner. Startups experiment more cheaply. Public services become more efficient. Companies that previously could not afford advanced AI gain access to it.

The benefits would appear across thousands of industries rather than only in the accounts of a single AI provider.

Even relatively modest productivity improvements, spread across an economy as large as the EU, could generate substantial value over time.

But we should be honest about the uncertainty.

AI adoption requires training, organisational changes and investment in complementary technologies. Public infrastructure could be poorly managed or become technologically outdated. The benefits might also be concentrated in certain sectors rather than distributed evenly.

A serious programme would therefore need measurable objectives, independent evaluation and the ability to redirect funding when projects fail.

The economic case deserves to be evaluated in terms of the productivity gains, strategic resilience and new opportunities it enables, not merely the revenue it generates.

A public AI system could be financially unprofitable while still being an excellent investment for European society.

That is already true of many forms of public infrastructure.

Wouldn't this destroy private AI companies?

I don't think so, and I wouldn't want it to.

The objective should not be to eliminate private AI providers or prevent American companies from operating in Europe.

Private companies would still have plenty of opportunities to compete by developing better models, offering specialised services, providing more computing capacity and building applications for particular industries.

A public model would establish a competitive baseline.

If a private company can offer a model that is substantially better, more reliable or more convenient, customers should be free to pay for it.

But businesses and citizens should not be forced into dependence on a small number of providers simply because no affordable alternative exists.

There is also a benefit for the private sector itself. Open infrastructure can make it easier for startups to build businesses without first raising enormous amounts of money to access foundational technology.

The result could be more competition, more experimentation and a larger ecosystem of European companies.

Public infrastructure and private innovation do not have to be opposing forces. Properly designed, they can reinforce one another.

What about the risk of government inefficiency?

This is a legitimate concern.

Governments are not necessarily good at developing technology products. Public projects can suffer from bureaucracy, political interference, slow procurement and poor execution.

An EU-funded model could also fall behind the frontier, particularly if it were managed as a political project rather than a serious research effort.

For this reason, I would not ask the European Commission to build a giant government AI company and expect it to outperform the private sector.

I would establish a public infrastructure programme with independent technical governance, transparent performance targets and multiple competing research teams.

Funding should depend on results. Models should be benchmarked against the best available alternatives, and projects that fail to deliver should not receive unlimited funding.

The EU should also work with existing European AI companies and research institutions rather than duplicating everything they do.

And because this infrastructure would be so important, its governance should include strong protections against political interference, independent audits and transparent rules governing access to public resources.

The public sector's role would be to fund strategically important capabilities, guarantee broad access and establish a competitive foundation.

It would not need to control every aspect of the ecosystem.

The answer to the risk of government inefficiency should be better institutions and accountability, not the assumption that foundational infrastructure can safely be left entirely to private interests.

What if the US achieves AGI first?

This is where my concern becomes more serious.

I don't know whether AGI will be achieved soon, what it will look like or which country will develop it first.

But imagine that one country develops AI systems capable of performing a very large share of economically valuable cognitive work.

Imagine that those systems can accelerate scientific research, develop software, design products, operate businesses and help create even more capable AI systems.

Their economic value could be extraordinary.

If those systems remain under the control of a small number of foreign companies, the consequences could extend far beyond software subscriptions.

The companies that own the systems could capture a substantial share of the economic value they generate. They could attract more capital and talent, reinvest their profits and accelerate their technological lead.

Countries that depend on these systems would still benefit from using them. But they might capture a smaller share of the profits and strategic advantages.

This could create a self-reinforcing cycle.

The leading companies would use AI to become more productive, invest more in research, attract more resources and improve their systems. Their customers would benefit, but the owners of the underlying technology could accumulate an increasingly powerful position.

For Europe, this could create a difficult situation.

Its businesses might become more productive while becoming increasingly dependent on foreign technology. Its governments might rely on infrastructure they do not control. Its startups might struggle to compete with companies that have direct access to the world's most capable AI systems.

And if AI eventually reduces the importance of human labour in producing economic value, ownership of the technology could become even more important than it is today.

If a growing share of economic output comes from AI systems, then who owns those systems, who receives their returns and who has access to their capabilities will become fundamental questions.

These are not merely technology questions. They concern the distribution of wealth, the future of work, the economic power of states and the ability of societies to determine their own direction.

I don't want Europe to find itself in that position without having invested in alternatives.

I am not arguing that the US will necessarily monopolise AGI, that Europe is destined to lose or that technological dependence inevitably leads to economic decline.

I am arguing that the potential consequences are sufficiently significant that Europe should not treat this as an ordinary commercial competition.

Europe doesn't necessarily need to win the AI race

I think this is where the discussion often goes wrong.

Europe does not necessarily need to build the single most powerful AI model in the world.

It needs to ensure that it can participate meaningfully in the AI economy, capture a reasonable share of the value generated and retain the ability to make independent technological decisions.

There is a difference between being the absolute leader and being technologically dependent.

Europe could benefit enormously from AI even if the most capable models were developed elsewhere. But it should have the infrastructure, expertise and alternatives required to avoid becoming entirely dependent on external providers.

Public AI infrastructure could be one part of that strategy.

It would also need to be accompanied by better access to growth capital, a more integrated European market, reliable energy supplies and policies that make it easier to create and scale technology companies.

Public AI alone would not solve Europe's competitiveness problems.

But it could provide a foundation on which many other solutions become more viable.

The question is not whether Europe can prevent every other country from advancing faster. It cannot.

The question is whether Europe will have enough technological capability, economic strength and institutional independence to shape its own future in a world transformed by AI.

The next Industrial Revolution is happening now

I don't see the USA as an enemy, and I don't think technological competition needs to become a zero-sum geopolitical conflict.

American AI companies have created enormous value, and European citizens and businesses should be able to benefit from their innovations.

But I also think Europe needs to be realistic about its long-term interests.

The Industrial Revolution changed far more than manufacturing. It transformed where people lived, how they worked, how economies grew, how wealth was accumulated and how power was distributed between nations.

AI could have a similarly broad impact.

It could reshape almost every industry, change the economics of intellectual work, accelerate scientific progress and alter the relationship between labour and capital.

If that happens, the countries that control and develop foundational AI capabilities will have opportunities that extend far beyond the technology sector itself.

Those that rely entirely on others may still prosper, but they will have less control over the forces shaping their prosperity.

History does not guarantee that technological leaders will remain ahead forever, or that those who fall behind can never catch up. But catching up can become much harder when technological advantages compound across an entire economy.

This is why I think Europe needs to take the issue seriously now, rather than waiting until AI has become indispensable and the most important capabilities are already concentrated elsewhere.

Instead of treating AI exclusively as a commercial product, the EU should consider treating foundational AI capabilities as a form of public infrastructure.

It could invest in computing capacity, support open-weight models, provide citizens with affordable access to capable AI and give European businesses the foundation they need to compete.

The investment would be substantial. The outcome would be uncertain. And the programme would need to be managed carefully to avoid wasting public money.

But the potential consequences of failing to invest could be much greater than the loss of a few technology companies or a single market opportunity.

If AI becomes one of the foundations of economic and societal life, Europe cannot afford to be merely a customer of the countries that control it.

It needs to be a participant in building that future.

The EU does not have to win every AI race. It does not have to own every model or prevent private companies from succeeding.

But it should ensure that its citizens, businesses and institutions have access to foundational AI capabilities, and that Europe retains meaningful control over the infrastructure on which its future prosperity may depend.

We made electricity broadly available because modern economies could not function without it. We built public infrastructure because its value extended beyond the private interests of any one company.

If AI becomes as fundamental to our economies as electricity, perhaps we should start thinking about it in the same way.

The next Industrial Revolution is not simply a technology race. It is a contest over the foundations of future prosperity and power.

If Europe wants to remain in control of its economic future, it should consider making AI a public good.

What do you think? Should the EU invest in public AI infrastructure and provide affordable access to European citizens and businesses, or would this be an expensive government project better left to the private sector?

I'd be interested to hear what other European developers think about this.