On 14 September in Vienna, the President of the European Central Bank, Christine Lagarde, warned that Europe’s dependence on foreign artificial intelligence systems could soon affect almost every part of the economy.
AI will be used to control goods at borders, select tax returns for audit, manage rail traffic, monitor patients and process bank payments.
Any change in price, technical conditions or access rights to such systems could affect several sectors at once. Lagarde linked this risk to the way European savings are invested.
European capital is abundantly financing US technology companies, while European firms find it difficult to raise the money needed to develop models and build computing infrastructure.
Eurozone households hold around €440 billion in shares of US technology companies.
The largest American companies in cloud computing and AI are increasingly borrowing in the European market as well.
According to ECB data as of 20 August 2026, the value of their bonds denominated in euros reached approximately €40 billion, and during 2026 they accounted for almost a tenth of new bond issues by non-financial companies in European currency.
Amazon and Alphabet were the two largest bond issuers among non-financial companies in the Eurozone market up to that date.
In March, Amazon sold bonds worth a total of €14.5 billion, the largest single corporate issue in the history of the euro-denominated bond market.
European households are participating in America’s tech boom through investment funds that buy shares, while pension funds, insurance companies and other large investors purchase the debt of companies that build data centres and acquire processors.
These investments generate income for European savers, but money from Europe finances the infrastructure from which most of the new technology, income and well-paid jobs will originate in the United States.
Capital flows to places where there are already large technology companies, developed stock markets and investors willing to wait years for returns.
US companies are also borrowing in euros
America’s largest technology companies have long funded most of their development from their own profits. Plans for new data centres have become so expensive that even companies with the highest cash flows are increasingly issuing bonds.
The Bank for International Settlements has calculated that more than $100 billion will be raised in the bond market in 2025, mostly with maturities of over five years, and that their capital investments could exceed a trillion dollars by 2028.
The public market covers only part of these costs. Tech companies form joint ventures with private equity funds and enter into long-term contracts to use data centres financed by their partners.
The liability is then kept off the technology company’s balance sheet, but the multi-year lease and guarantees have almost the same economic effect as debt.
Such deals are accelerating construction and drawing banks, insurers and private equity funds deep into financing the AI industry.
Demand for European corporate debt has remained strong
The European corporate bond market is attractive to US companies because it provides access to additional investors and often a lower cost of borrowing.
It absorbed several very large issues during 2026 without difficulty, but the biggest users of that capacity are research centres, major infrastructure projects and much of the future business activity in the United States.
So far, the ECB does not consider that the large US bond issues have made it difficult for European companies to sell their own bonds.
Demand for European corporate debt has remained strong, although some firms have postponed sales to avoid direct competition with US technology groups for the same investor funds.
The growth in this borrowing has already led investors to seek higher yields. Lagarde warns that the continuation of such borrowing could raise long-term interest rates in the eurozone and thus increase financing costs for European companies. The ECB has not seen such an impact yet, but the risk increases with each new large issue.
European austerity bypasses domestic technology companies
European households save approximately €1.4 trillion a year, but the financial system channels only a small portion of this into the equity of companies that develop expensive models and build infrastructure for years before achieving stable income.
Much of the money remains in bank accounts, in domestic funds with conservative investment rules, or flows to more developed markets outside the European Union.
European companies still rely heavily on bank loans. Banks are well suited to financing businesses with assets, contracts and predictable revenues.
Developing the most advanced AI models requires capital that can withstand years of losses and the possibility of total failure. In the United States, that risk is shared by stock markets, venture capital funds, private lenders, insurers and the bond market.
A company that wants to raise capital across the European Union must navigate differing tax rules
A company that wants to raise capital across the European Union must navigate differing tax rules, investment fund regulations and financial supervision systems, as well as varying recovery procedures in the event of insolvency.
It is therefore easier for young companies to seek their next major investment in London, New York or Asia than in several European countries at the same time.
French company Mistral raised €3 billion at the beginning of September, valuing the business at around €21 billion. It was the largest equity financing round for a private European technology company.
Mistral, however, is an exception. Developing the AI industry requires consistent funding for a wide range of companies, from model developers to operators building data centres.
The round was co-led by PSG Equity, Samsung Electronics and the EU-backed fund Scaleup Europe, showing that even Europe’s biggest AI projects still rely on capital from outside the EU.
America has three quarters of the world’s AI computing power
US tech companies have already turned their funding advantage into chip factories, data centres and computing power.
The United States has approximately 75 per cent of the world’s capacity for AI development, while Europe accounts for about five per cent.
In 2025, Stanford’s AI Index recorded 59 models developed in the United States and 35 in China, while France and the United Kingdom had one each.
This category includes models that have achieved leading results in recognised tests, reached widespread use, attracted a large number of professional citations, or cost more than a million dollars to train.
These data measure the extent to which each country has developed models that have reached a certain level of technical or market significance.
Greater computing capacity allows American companies to develop multiple systems simultaneously, test different technological solutions and absorb the costs of failed projects more easily.
In June, the European Commission proposed the Cloud and AI Development Act, which aims to triple data centre capacity in the European Union over the next five to seven years, accelerate the granting of permits and expand the range of services over which European institutions have greater legal and operational control.
An estimate prepared alongside the proposal shows that the gap between demand and available capacity could grow from approximately three gigawatts in 2025 to around 20 gigawatts by 2036.
Large computing centres for developing and training AI models can be launched with public funding
According to this estimate, presented by Lagarde, closing that shortfall could require up to €600 billion, including the cost of chips. The ECB warns that the actual cost is likely to be lower because the calculation is based on the most expensive type of AI data centre.
Large computing centres for developing and training AI models, which the European Commission refers to as AI gigafactories, can be launched with public funding.
The state can cover part of the construction and research costs or pre-contract the use of computing capacity, thereby reducing the initial risk for private investors.
National budgets, however, cannot finance the entire industry in the long term at the scale that Europe needs.
Greater private investment will require cheaper and more reliable electricity, faster permitting, a secure supply of advanced processors and sufficient customers across the European Union.
If these conditions are not met, European centres will be built more slowly and will have less computing capacity than American ones.
European data and public services on foreign infrastructure
Buying a foreign AI system is fundamentally different from buying an industrial machine. With most widely used commercial systems, the model and core infrastructure remain under the control of the supplier.
The system is constantly evolving, and access depends on the contract, price, technical conditions and the regulations of the country whose jurisdiction the company falls under.
European companies often entrust such systems with data on production, customers, business processes and research.
Contracts can prohibit the use of that data to train models and thus reduce commercial risk, but access to the technology still depends on a foreign supplier and a foreign legal framework.
Disruption originating in the United States is currently unlikely, and European companies and institutions use American systems because they are available, advanced and often cheaper than European alternatives.
However, the terms may change as a result of new export restrictions, sanctions, disputes over digital taxes or suppliers’ commercial decisions.
When the same systems are used in tax administration, healthcare, energy, transport and banking, a change of approach no longer affects only the company that bought the software. The consequences extend to public services and sectors of the economy that depend on those systems.
Europe is likely to continue using American models across much of the commercial economy, where their performance and price offer a clear advantage.
At the same time, it will invest in domestic infrastructure for state administration, defence, energy, health and other sectors that handle sensitive data.
This approach is financially and technologically more feasible than attempting to replace all American services in a short period. By the end of the decade, a European system could be in place for the most sensitive state and industrial needs, alongside the widespread use of American technology in other parts of the economy.
AI is putting capital markets back at the top of the European agenda
In the same speech, Lagarde linked Europe’s technological backwardness to its still incomplete single capital market.
For years, the Savings and Investments Union has been confined to technical discussions about stock exchanges, investment funds, insolvency rules, taxation and cross-border supervision.
The development of artificial intelligence has given that project immediate economic and security relevance. Europe can hardly control the infrastructure on which it relies if its companies do not have access to the capital needed to build it.
Mistral is currently the strongest candidate to become Europe’s leading manufacturer of advanced models and data centres dedicated to public services, defence, healthcare and industry
Over the next two years, Brussels will seek to channel most pension and investment fund money towards European technology companies, facilitate investments across national borders and reduce the costs that differing rules impose on young firms.
Resistance will come from governments that want to retain control over tax rules, pension systems and financial market oversight. Progress is therefore likely to be slower than the growth of US data centre investment.
By 2030, Europe may have several domestic manufacturers of advanced models and data centres dedicated to public services, defence, healthcare and industry.
This would provide a reliable European option for storing sensitive data and operating key systems.
Mistral is currently the strongest candidate to become Europe’s leading manufacturer of such models. Its position will depend on its ability to secure capital for each subsequent generation of technology, not just for the current investment round.
If governments again delay capital market integration, European savings will continue to flow to companies that already have large equity markets, developed bond markets and investors willing to wait years for returns.
European institutions and companies will then buy AI services from American firms whose development was financed by European citizens, while domestic companies lack the capital to expand beyond national markets.
European savings can fund domestic computing infrastructure only when markets are able to take large, long-term risks. Otherwise, most of the value generated by AI will be created outside Europe, even though part of the bill is paid by European investors.
