Attention is all you need – why Europe should decouple from US technology

In a landmark 2017 research paper, eight scientists proposed novel architecture for improving machine learning design, which has since become one of the pivotal breakthroughs for modern Artificial Intelligence (AI).

Their work detailed an essential component of the large language models (LLMs) used by services like ChatGPT, namely the “attention mechanism” and advanced “weights” that calibrate the significance of input data to produce the best output. As a homage to the Beatles, the paper was wittingly titled “Attention is All You Need”, which not only encapsulates the most important raw component in today’s AI design but also, ironically, the state of our whole digital ecosystem.

Attention as currency

Though few of us realise it when we browse a website, online advertising relies on a pervasive system known as “real-time bidding”. This effectively functions like an ultra-fast online auction in which advertisers submit bids in the milliseconds it takes for a page to load, ensuring that users get the most applicable advert based on their preferences and individual search history.

What is even less well known is that the data points and intimate insights fuelling real-time bidding are freely and legally sold by data brokers in bulk. This model is fully acknowledged and tacitly supported by the US government and its security services. Analysing these data points and patterns improves American surveillance capacity globally, while data brokers and attention merchants have a lucrative incentive to continue the model.

The emergence of AI threatens to further complicate this picture. US Big Tech companies are already investing more than $300 billion in AI spending for 2025 alone, while total global investment in data centres and AI-related infrastructure is predicted to reach a staggering $1.5 trillion.

Global online advertising is dominated by just three American companies: Google (Alphabet), Meta and Amazon. The emergence of AI is likely to follow a similar pattern, with a handful of American companies (Open AI/Microsoft, Meta, xAI, Anthropic and Google) now vying to establish monopoly status in this space. Their goal is clear – dominate this market and become embedded in the phones and lives of billions of people.

The societal challenge posed by AI

Linus Torvalds, the creator of Linux, has stated he believes today’s AI industry is 90% marketing and 10% reality. Other computer scientists and academics have characterised AI as a “con act” and warned of “AI snake oil sellers”. Indeed, the term “artificial intelligence” is now being applied in such a broad way that it can mean everything and nothing.

What is clear, however, is that AI could quickly overshadow many of the current incumbents in the field of online search, social media and the app economy. Traditional search engines are already losing ground in terms of online traffic as LLMs suck up queries and redirect users to news, platforms and services.

The most established LLM companies will most likely concentrate a huge amount of wealth, personal data and computational resources. More worryingly, future AI assistants will have direct sway over how people make up their minds about personal actions, purchases and even political choices.

Researchers already warn that advanced chatbots and AI companions are opening a new frontier of persuasive technologies with access to intimate thoughts, emotions and personal patterns. It seems just a matter of time before we see advertising directly ingrained in new AI models. This would be the logical conclusion to the attention/surveillance capitalism that dominates our digital ecosystem.

Yet the consequences of AI failing could be similarly disastrous. The whole industry might come crashing down in the short to medium-term. Close to a third of today’s stock market is tied to a handful of US tech companies burning tons of cash with little actual revenue in return. A recent WSJ deep dive highlights the abysmal finance model behind AI investment and data centres. This means the AI boom could quickly turn into a major bust.

The EU needs a better alternative

Regardless of the direction of travel for the AI industry, the EU is far from the driving seat. Europe is handicapped when it comes to venture capital and its overall attractiveness for high-risk investment. But the biggest problem is that the onslaught of US and Chinese software/hardware has almost gobbled up the whole European digital ecosystem.

A staggering 80% of Europe’s digital infrastructure, software and services are imported from abroad. This has serious national security implications due to over-dependence on third-country providers, while a much needed “European Way” in navigating a sovereign Eurostack, infrastructure and services is still missing.

As it stands, Europeans will pay perpetual rent to foreign businesses for the services they are dependent upon for work and leisure. At the same time, their digital landlords have rights on their personal data, with the ability to carry out surveillance on European citizens and weaponise the increasing dependencies.

All of this comes with the protection of the US government. We have already seen the current Trump Administration strong arm the EU to adapt its own digital rules and tax regime to serve the interests of Big Tech. We now live in a world where a presidential executive order can directly suspend access to personal mail accounts or Microsoft services.

Decoupling from the US

If Europe is serious about tackling these issues, it needs to jump-start its competitiveness by curbing unnecessary legislation that has created a surreal economy of costly compliance. An urgent push is also needed to create higher internal demand for European services and unrestricted capital movement within the Union.

When it comes to money, the current EU budget offers huge amounts of funds that are riddled with inefficiencies, redundancies and outright fraud. Each year the EU allocates more than 100 billion euros on cohesion and agricultural spending which yields less and less added value.

A true reformist pivot would also mean re-evaluating the way we are conducting our climate and energy policy. The EU cannot honestly expect to have a successful twin digital/green transition simultaneously with the current status quo.

An economic model set up on embedded high energy costs, a shortage of energy supply and ever-increasing climate mandates has little chance to scale the needed capital and trust. Not to mention digital infrastructure, quantum computing or next-gen technologies, which would require ever-expanding energy resources.

There is also an important case to be made about the diffusion of technology and AI breakthroughs into the wider European economy. Technological advancement is not about who has the best breakthrough model or number of STEM graduates, but how new tech and related productivity gains are integrated and adopted among users, businesses and government agencies.

Europe actually has the labour force and some of the essential pre-conditions for becoming a smart second mover in the AI race. Regrettably, however, the pattern of the past is just being repeated with LLMs. Similar to social media and digital services, the EU is likely to follow in the footsteps of the Big Tech monopolists developing AI and become increasingly dependent on them.

Given recent geopolitical developments, it is clear the only viable option is for the EU to try to decouple from reliance on the US for essential digital services and infrastructure. Fully decoupling on all digital fronts is not realistic, but there needs to be a targeted effort in core digital domains.

A safe and sovereign “Euro cloud”, increased targeted investment in data centres, a dedicated EU Tech Fund and citizen services that are run on European servers should be the norm. The European Commission needs to push hard on ensuring interoperability and avoiding the costly mistakes of social media design where users and their data are locked in a digital vassalage.

Technological sovereignty is a welcome rallying cry, but it is about time for this mantra to receive actual political backing and a clear strategy behind it. All of the AI Acts and GDPRs will bring Europe nowhere unless it has the infrastructure and independent capabilities to preserve and propel its own interests.

Note: This article gives the views of the author, not the position of EUROPP – European Politics and Policy or the London School of Economics. Featured image credit: Antonello Marangi / Shutterstock.com