The EU AI Continent Action Plan: Hype, Burn, Rinse, and Repeat

The European Commission’s AI Continent Action Plan is closer to fiction than reality. Steeped in hype, the EU will burn more fossil fuels, rinse and repeat the past, all in the name of leadership.

The Commission’s Action Plan, launched this Wednesday, has five elements: more computing infrastructure, more access to data, more adoption of AI systems, a larger talent pool of AI experts, and a large single market. Just as the EU—unlike Europe—is not a continent, this Action Plan is not a plan. While the Commission recognizes the EU’s dependency on non-EU infrastructure, its “plan” still apes OpenAI’s blueprint.

Hype

The Commission’s Action Plan is built around the speculation of a “transformational shift” shaped by using more data centers and more data for AI. This approach is not new and has been embedded in the Communication on Fostering a European approach to Artificial Intelligence since at least 2021.

What is new? The Commission’s wholesale acceptance that “the next generation of frontier AI models [are] expected to unlock a leap in capabilities, towards Artificial General Intelligence.” The Commission has bought into the hype of “artificial general intelligence (AGI)” and “frontier models”—two terms promoted by OpenAI, Anthropic, and the AI-as-existential-risk troop. The term “frontier model”, a rebranding of generative AI models, has been popularized by the Frontier Model Forum, which counts Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI as its members.

Reflecting Big Tech’s “more data is good” ideology, the Commission will establish a new surveillance infrastructure called Data Labs, “to enable the provision, pooling, and secure sharing of high-quality data.” Data Labs “ensure that AI developers will have access to large volumes of high-quality data in health” from across the EU. Data Labs will create a database of ruin, take control away from people, and their privacy. Quoting Paul Ohm’s piece as it is still relevant today, “Once we have created this database, it is unlikely we will ever be able to tear it apart.”

Tech CEOs are hallucinating about the capabilities and benefits of these “frontier models” while propagating harms. Now, the Commission joins them in this hallucinatory experience—all at the expense of us and the planet.

Burn

This hype-based policy shows that the Commission has been gaslighted into believing it will become a leader if it follows Big Tech’s approach.

The Commission hides behind superlatives such as “AI Gigafactories”, “unprecedented scale”, and “hundreds of trillions of parameters” to describe what will lead to increased energy and water usage. Under the plan, the Commission will support a network of data centers to “integrate massive computing power, exceeding 100,000 advanced AI processors” to train AI models.

The Commission’s (lack of) imagination reveals itself when it indicates that all AI models are trained with “vast amounts of data and requir[ing] massive computational resources”, then optimized for specific applications before being integrated into applications. This is indeed the strategy of Big Tech: to train the general models and keep customers reliant on their cloud infrastructure. However, this is not the only approach. Frugal AI approaches use less computation and less but high-quality contextual data to build AI applications directly. The Commission, in its plan, makes no mention of such approaches to AI.

The advanced AI processors, most likely to be procured from Nvidia, to power the data centers will not only increase training and inference efficiency but also increase energy use and waste heat production. As a result, water required to cool the data centers will increase. Although “energy, water efficiency and circularity” should be taken into account, it is not clear what “taking into account” means when the plan includes “tripling the EU’s data centre capacity within the next five to seven years.”

By building and using more data centers for AI, Ireland burns more fossil fuel while Big Tech depletes water resources in drought-affected regions. The EU’s policy remains on course to combine the worst of Ireland and Big Tech.

Rinse and repeat

The Commission's plan is not just to have more data centers and more data approaches, but also to repeat past announcements. The EU has already planned to build 13 AI factories, which are to be equipped with new or updated supercomputers and data centers, where researchers can test and develop AI models. These “factories” include consortiums involving some of the leading AI research groups in the EU, many of whom work on a wide range of AI techniques and applications that are not the Big Tech approach, which the Commission parrots.

While the Commission mentions “open innovation” and acknowledges “AI model development in the EU benefits from advances in open-source approaches,” it does not state that open-source projects will receive priority access to the supercomputers. European users, including “startups, scaleups, SMEs”, will have priority access through EuroHPC. “European” users broadly cover users in the EU and “in a Participating State or in a third country associated to the Digital Europe Programme or to Horizon Europe.” This would include the UK, Canada, and New Zealand.

What about the EU? It is home to more than 6,800 AI startups,” according to the AI Continent plan. For this number, the Commission relies on a report that covers generative AI, as if other AI startups do not matter. The report itself can only be accessed after sharing personal data. True to the AI hype, the Commission provides a non-existent link to that report, as if generative AI produced it.

Elaborating on the previous announcement to set up a European AI Research Council, the Commission will support “Science for AI” and “AI in Science” to foster “the use of AI for discovery and exploration across a range of scientific disciplines, unlocking cross-pollinations between AI and domain sciences.” Without “clear scientific guidelines on how to use” AI in specific domains, this approach could be more harmful than useful. In the words of two computer scientists, “Science is not merely a collection of facts or findings. Actual scientific progress happens through theories, which explain a collection of findings, and paradigms.”

The Commission assumes that more deployment of AI systems will produce novel solutions and improve public services. It worries that “only 13.5% of companies in the EU had adopted AI,” as if other companies are missing out on something good. But, “[t]here is nothing that makes AI systems inherently good.” Rinse and repeat of more data centers and more data ideology will not change that.