Europe has missed out on multiple technology-driven booms over the past three decades, which has been a key factor in its sluggish economic performance. As artificial intelligence (AI) emerges as the next big technology wave, it cannot afford to miss the opportunity. A key question is what strategy should the European Union (EU) pursue to secure its place in the global AI ecosystem?
Some European governments and companies believe that to make this happen, they need to develop low-cost copies of ChatGPT in local languages for their local markets. This is a mistake. A more effective approach to developing AI in Europe might be to establish a CERN for AI, develop more domain-specific AI systems, and create a robust AI assurance industry.
Europe's economy needs a boost. An ageing population, a worsening climate crisis, a lack of affordable housing, high energy costs, and many other factors are weighing on already strained national budgets. The exact value added that AI will bring to the global economy is unclear, but its size and growth could be substantial. PwC predicts that its potential economic contribution to the global economy could grow to $15.7 trillion by 2030, equivalent to the combined GDP of the EU's 10 largest economies.
The European Council has recognized the enormous economic potential of AI and called for strengthening European capabilities in this “key technology of the future,” a sentiment echoed by countless thought leaders, from CEOs to politicians to academics. But the value AI can bring to the European economy depends on a strategy that ensures the European Union has a strong position in the global AI ecosystem.
Half-heartedly pursuing the European GPAI development will lead to failure
Too often, European companies and governments are making superficial efforts to develop general-purpose AI (GPAI) in local languages. This enthusiasm is understandable: GPAI systems can respond to simple questions and assist with a variety of everyday tasks. Who wouldn’t want support in their native language? But the trend in GPAI development investment is that bigger is better, so these efforts may lack the funding to compete with current market leaders. Google’s latest Gemini Ultra model cost a staggering $191 million to train, contrasting with European efforts such as the Dutch government’s €13.5 million investment in Dutch GPT. Moreover, the enormous development costs are becoming increasingly difficult to justify on linguistic grounds alone, given that ChatGPT, for example, already supports 22 of the 24 official EU languages.
As a result, trying to compete with American AI systems and similar GPAI models will be an infeasible strategy for Europe without unprecedented levels of government support. According to the latest estimates, to maintain the current growth in GPAI size, global AI chip production would need to increase 25-fold by 2030. In addition, global electricity production would need to increase by an amount equivalent to 25% of Europe's energy use each year.
In other words, the EU must either go big or go home. The scale of investment is too big for any European company or country to manage. As the Commissioner correctly pointed out, only a resource pooling effort by a European AI research council like CERN could mobilize the necessary capital. Yet to successfully develop GPAI, the EU will need fast-track investment, a single market, and reform of insolvency rules. In short, the EU needs to consider whether it can create the conditions to play in the AI big leagues based on infrastructure, financial, and political factors.
Two underrated European strengths for securing AI market share
Instead of pursuing only general-purpose AI systems, the EU should consider diversifying its innovation portfolio by investing in another kind of powerful AI: domain-specific systems. In certain highly specialized tech industries with high-quality proprietary data, these narrower scope models often outperform GPAIs in niche but important tasks. The benefits can change the lives of millions of people. For example, domain-specific AI models are already accelerating vaccine development. Moreover, domain-specific AI systems are not only more effective than GPAIs in their niches, but they are also significantly cheaper to train. Alphafold3, the flagship domain-specific AI system for life sciences, cost just $1 million to train, 100 times cheaper than GPT4.
Fortunately, the EU can leverage existing strengths to develop state-of-the-art domain-specific AI in key industries. A unique development challenge for domain-specific AI is finding large and diverse training and testing datasets in the respective domain. Several European companies, such as Novo Nordisk in life sciences and SAP in enterprise software, are already at the forefront of their respective domains, with their own high-quality data.
If European leaders have the computing access and talent they need for AI development, it could help consolidate their global industry leadership. Moreover, if the upcoming European Data Integration Strategy announced by European Commission President von der Leyen achieves its goal of streamlining data sharing, other European companies will be able to benefit from the wealth of domain-specific data generated by European market leaders.
Moreover, the EU is uniquely well-positioned to develop a burgeoning AI assurance industry. Companies that provide software, hardware, and services to help other companies build and deploy well-maintained, legally compliant AI systems have a fast-growing niche. AI assurance companies have already secured a €1 billion market globally, and market growth is expected to reach hundreds of billions of euros globally by 2030. EU companies operating under the world's first binding international AI treaty can leverage their experience in AI risk management to develop audit, risk advisory, and other assurance tools. European AI assurance companies also have other key assets that could prove valuable in the industry. Access to an agile, world-class AI talent pool, a strong traditional audit industry, and top research universities provides aspiring EU assurance companies with the talent and avenues for intra- and inter-sector collaboration.
Europe still has time to change its approach to ensure its competitiveness in global AI developments. But it must stop wasting resources on half-hearted, piecemeal efforts. Instead, the continent must build on its strengths and focus efforts on jointly developing competitive GPAI models, domain-specific AI systems, and a world-leading AI assurance industry.
