Making Sense of Europe's AI Investment Plans
Last week Macron announced €109bn and Von der Leyen announced €200bn for AI. What does this mean? What is it for and will it work?
A few weeks after Trump announced $500 billion "Stargate" investment in AI (a real number? who knows!), the French government and the European Commission have announced €109bn and €200bn respectively. The UK launched their plan in January, but they didn't have a number of billions attached.
How do we assess these announcements? The bigger the better? What are they trying to achieve and how likely are they to work? Let's break down the announcements and ask some key questions:
Why is investment in AI seen as so important?
I think there are two primary drivers: productivity growth and military capabilities. Mario Draghi's analysis in his famous report specifically says the productivity gap that has developed between the US and the EU over the last two decades has been driven almost exclusively by the US tech industry:
The EU economy has traditionally been strong in all mid-technology sectors that are not at the centre of radical technological advances.[...] The EU has less activity in sectors in which much of the productivity growth has originated in recent years, notably the ICT sector and the exploitation of large-scale digital services. Due to slow technology diffusion within industries, the EU’s productivity growth gap compared to the US was particularly pronounced in these industries with very high productivity growth. [...] Excluding the main ICT sectors (the manufacturing of computers and electronics and information and communication activities) from the analysis, EU productivity has been broadly at par with the US in the period 2000-2019. For 2013-2019 the role of ICT is even more striking, as the EU productivity growth excluding the main ICT sectors exceeded that of the US by some margin
Europe missed the last wave of rapid productivity at the technological forefront. AI looks to be a very likely candidate for the next wave, so European leaders are determined to avoid missing this one.
AI is also very likely to be a significant component of future military capability. From autonomous drones to cyber security to potentially even strategic decision making. A slight lag behind the US might not have been a huge concern, especially if we can buy tech from them, but the Trump administration makes this a more precarious position with each passing week.
This is an economics newsletter, and I'm not a military expert, so I'm going to focus on the productivity implications.
Two Paths to AI Prosperity
There are two ways Europe could benefit from the AI revolution:
The Safe Path: AI-ification This is about making existing industries more productive – helping lawyers draft faster, creatives generate more ideas, pharma companies discover new compounds. Think of it as this century's version of electrification or digitization. But while important, this isn't where the real race is happening.
The Bold Path: Owning the Future This is the big bet: ensuring the core, leading-edge AI companies are European, so the value accrues here. As Draghi mentioned above, we have had a bad experience of not "owning the future" when it came to digitisation. Our businesses succeeded in digitising by using Microsoft Office and advertising on Google and selling their products on Amazon, which has increased their productivity, but a huge portion of the gains they make are then getting captured by the US.
That’s what we don’t want to see repeated on AI, but it’s not quite clear yet exactly what “owning the future” will look like.
What might success look like?
In 1995 most industry analysts would have guessed that owning the browser was key to owning the future of the internet (and Netscape and Microsoft fought huge battles over it), but Europe never had a major browser* and that was mostly fine. It turns out that owning the aggregators was key, but nobody figured that out until much later.
Let's look at four potential layers of abstraction, each of which has the potential to be the most important one to own.
The Foundational Models: Inventing the Turbine
The forefront of AI science, at the moment, is in Large Language Models. In particular, much of the excitement and attention is focused on so-called "Foundational Models". These are the big engines at the core of the new wave of Generative AI. They are the electricity turbines. The biggest ones are as follows:
US dominates with Claude (Anthropic), Gemini (Google), GPT (OpenAI), and Llama (Meta)
China competes with Deepseek and Qwen (Alibaba)
Europe has one, Mistral
The good news? We have Mistral, and it's genuinely competitive. The sobering news? Mistral's €30 million in revenue looks tiny next to OpenAI's estimated $3.5 billion, or even Anthropic’s $400m.
There is a version of the future where owning the foundational models might be key to owning the AI future. If that's the case, Mistral is our best bet, but the mountain ahead of us is a very steep one.
But to be honest, I don’t think that’s the most likely version of the future. It might have seemed that way a year or two ago, but as things have progressed we have seen a wide array of foundational models get developed, each with very little differentiation and all starting to plateau in terms of value.
As analyst Benedict Evans said recently "Google launched v2 of its Gemini LLM route. Back to the top of the benchmarks. Well done, but also, so what? At this point we know that there'll be lots of these, they'll be mostly a commodity, and what matters is the product you built on top."
Deepseek's recent launches show that model distillation might also be a viable strategy. This means that after the leading companies launch their models, fast-followers can use those models to "teach" cheaper and smaller offspring.
New forks in the road like deep research, multi-step models and agents might bring this layer back into prominence, but it’s not yet clear that you need to own the foundational models to win at any of these, or that they will be big winners (although deep research seems like a good candidate for both).
The Application Layer: Light Bulbs and Kettles
So let's look at "the products you build on top". These are the applications which will be powered by the foundation models. The first will be the pure-play LLM services, the broad purpose, consumer facing apps like Chat GPT. This is a place where I worry that Europe will not be able to compete.
The dynamic here is likely to have aggregation effects. The largest chatbots will have the most users, which will help them to learn more, improve and personalise their services better.
But there will also be many, many other new applications above and beyond the chatbots. Marketing and customer service and legal services and image editing and copywriting and legal drafting and diagnostics and software coding and dozens of other industries are ripe for disruption.
This feels like the most promising place where Europe can own a slice of the future, if we double down and move fast.
Much like the transition from desktop to mobile, the move to AI will represent a window of opportunity for a new set of winners to emerge. I don't know if this will definitely be the space where the future gets owned, but I do believe it's Europe's best angle of attack.
We should do everything we can to make this the version of the AI future that unfolds.
The alternative future is that the application layer could evolve mostly through big and existing companies integrating AI into their existing services. This will also most likely benefit the US tech companies (Microsoft word and Google docs will write your notes, Gmail will compose your emails etc.). There will also be opportunities for some of the large European players to do the same. Spotify can make better playlists (and maybe audio ads), pharma can discover new compounds and Allianz can make better predictions around insurance and Accenture can help everyone implement chatbots. But for many of the large European companies (think LVMH, Hermès) the opportunity isn't really there.
If that's how the application layer evolves, the US feels like it has a significant edge here.
Infrastructure: The AI Factories
AI powered applications will need to run on computers, and it's good business to cluster those computers together and rent out their capacity. This is what most of the French funding announcement was about. The vast majority of Macron's package consisted of private investments in the development of new data centres. About €20bn of the European Commission's package was funding for "AI Gigafactories".
Again this feels like a very difficult space for the EU to catch up. The big US cloud builders are going to spend about $300bn next year to further build out their data centre capacity. Europe's largest tech company, SAP, is actually lowering their Capex spend each year, to just under €1bn in 2024.
The current data centre industry has massive economies of scale from global distribution. There is no benefit for a digital service to host their app on “local” infrastructure, when the goal is to make it available globally.
That dynamic might change a bit for AI, if hugely costly model training remains a bit part of the future. In that case, access to cheap and plentiful power generation (like France’s nuclear cluster) may be a strategic differentiator.
Even then, all you’re really selling is electricity. Data centres don’t tend to come with many high paying jobs attached. The UK’s AI Action Plan was accompanied by an announcement of a £12bn data centre investment and 11,500 jobs. As Liz Carolan noted “That is one (yes one) job per approx. £1,000,000 of investment.”
The Draghi report echoes this sentiment:
"The EU's competitive disadvantage will likely widen in cloud computing, as the market is characterised by continuous massive investments, economies of scale and multiple services offered by a single provider.
The Commission’s announcement does include a focus on building 7 new “AI Gigafactories” for use by scientists, researchers and universities (amongts others) in a CERN style public-private partnership. Draghi also highlights other reasons we may want to play here, even if we can’t win:
it is important that EU companies maintain a foothold in areas where technological sovereignty is required, such as security and encryption ("sovereign cloud" solutions)."
This seems fine to me, if we maintain honesty about the goals and potential benefits. The EC's announcement unfortunately also pitches the Gigafactories as something that "will enable all our [..] companies – not just the biggest - to develop the most advanced very large models needed to make Europe an AI continent."
This makes me worried. There is plenty of choice in the cloud market already. If you really wanted to achieve this goal you'd probably be better off handing out a few billion worth of AWS credits.
So having a few "AI factories" of our own won't help us own the future by building hyperscale cloud companies, but it might be a good redundancy for governments and military and crucially the scientify and research angles could be an important driver for the 4th and final dimension - talent.
Growing the talent
No version of owning the future is possible without growing, attracting and retaining the right talent. Here neither of the funding announcements go into great detail, but they do both seem like decent precursors and enablers to attract good talent.
Overall Europe has a very high volume of academics working in this space, but we underperform when it comes to growing and attracting the very best of the best.
Will it work?
This €309bn of new funding is great to see. The French one is mostly private funding focused on data centre build out which, while not sufficient for owning the future, is exactly the right country in which to be building out this capacity.
The European Commission's funding announcement consists of €50bn of their own money, paired with €150bn of private capital. €20bn of this is for their AI gigafactories and it's unclear what exactly the rest is for. Most of the private investment seems to be focused on the application layer, which is great, but the Commission's investments here need to be much more focused. It talks of supporting SMEs and a focus on safe and Trustworthy AI, which risks becoming a bit of an everything-bagel approach.
We need an even sharper focus on winning the race to build the dominant, highest productivity, value generating players at the application layer while attracting and growing the best talent. This can be supplemented with defensive strategies in foundation models and data centres. Everything else is icing on the cake.
The most obvious place where Europe can win is a) in infusing new AI technologies into our existing corporate giants and b) investing in a new wave of AI-powered, high productivity startups which will take advantage of this once in a generation opening.
If you found this analysis valuable, consider subscribing to European Abundance Agenda for more insights on Europe's technological and economic future.
* The Opera browser is a spin-off from a Norwegian telco, but it peaked at 6.5% market share.