Commentary by Oisín Cahalane
In February, Anthropic, widely considered the developer of the world’s most capable AI, was blacklisted by the US government after refusing to comply with Pentagon demands. Within hours, OpenAI filled their position. Washington made its position unmistakably clear: companies that do not align with their desires will be easily substituted.
Europe, which does not have its own frontier AI model, is not ambitious enough to seize this moment. It is trailing the USA and China in the AI race by a considerable margin. What should have been done is this: Europe should have offered Anthropic CEO Dario Amodei to base his company across the Atlantic. Amodei and Anthropic have just seen how quickly the Trump administration can turn on a firm that refuses to compromise its principles. Europe offers something the US currently cannot: political stability and a regulatory culture far more aligned with Anthropic’s values around AI safety.
A deal would benefit both parties. A continent that cannot access frontier AI on its own terms is a continent that has ceded a defining technology of the 21st century to others. The same way defence dependency has left Europe exposed, AI dependency carries similar autonomy risks. Anchoring the world’s leading AI company in Europe would be a statement of intent, making the region a genuine player in the AI race.
The Defence Parallel
Anthropic’s Claude is widely agreed to be one of the most advanced AI chatbots. Foto: Wikimedia Commons->https://commons.wikimedia.org/wiki/File:Claude_AI_logo.svg] | 人人生來平等 | Public domain
For decades, Europe assumed the US security umbrella was a permanent fixture. However, politically conditional and delayed military support from Donald Trump during the war in Ukraine demonstrated that the US is no longer a reliable defence partner. Concerns were initially raised when Trump pressured NATO members to commit five percent of their country’s economic output to defence, and trust was shattered when the president said he wanted to take Greenland “one way or another”. Europe is now scrambling to rearm and prove it can independently resist foreign hostility.
“NATO WASN’T THERE WHEN WE NEEDED THEM, AND THEY WON’T BE THERE IF WE NEED THEM AGAIN. REMEMBER GREENLAND, THAT BIG, POORLY RUN, PIECE OF ICE!!!” - President Donald J. Trump pic.twitter.com/xgEV8P1n4n
— The White House (@WhiteHouse) April 8, 2026
The AI situation is structurally identical to that of defence: independence in the field is crucial for autonomy. We are living in a world where artificial intelligence is already determining economic power, national security, and information control. Strategic autonomy in AI is no longer optional but essential for any global actor seeking to maintain sovereignty and influence. Europe has world-class researchers and institutions, but has allowed the actual infrastructure of frontier AI (the compute, the chips, the labs) to concentrate almost entirely in America. The risk of being dependent in such a crucial field proves that Europe has not learned any lessons from Trump’s and America’s unreliability as an ally.
The Opportunity
Trump’s public feud with Anthropic has created a rare opening. Political interference has burned one of the world’s leading AI companies; who now has to fight to protect its principles. A move to Europe is a win for both parties. Dario Amodei’s company can be based in a region where its values are respected, and Europe becomes a legitimate figure in the AI race.
The Hard Reality: Infrastructure and Chips
On paper, the deal sounds great. Unfortunately, Anthropic cannot simply be packed into a box and shipped across the Atlantic overnight. If Europe wishes to get serious about AI, it needs to get serious about its infrastructure.
Immediately, Amodei would require two gigawatts of data centre energy to operate at frontier level. Europe only has around 15 gigawatts of data centre capacity compared to 50 gigawatts in the US, and this gap is only expected to widen in the near future. While new capacity is being developed under initiatives like the UK’s AI Growth Zones, they are not expected to be fully operational until 2030. Just as Europe could not rebuild its defence strategy within a year, it cannot conjure frontier AI infrastructure quickly. The continent is once again staring at the bottleneck of energy, planning reform and political will, and it is not responding with urgency.
Even with data centre infrastructure in place, Europe faces a semiconductor dependency problem that mirrors its energy dependency from Russia. These highly specialised processors, critical to frontier AI, are monopolised by Nvidia, an American company. The US has used semiconductor export controls as a diplomatic weapon across multiple administrations, and a “Europeanised” Anthropic could be cut off from the Nvidia chips and cloud infrastructure it depends on.
Act Now or Fall Further Behind
Europe will not reinvent frontier AI from scratch, but it must create conditions where it is not entirely dependent on American political stability to access it. What this means is serious coordinated investment in compute infrastructure across the UK and EU, and calculated offers to frontier labs.
The Anthropic window will not stay open for long. In years to come, this may feel like an opportunity squandered by the same Europe that sleepwalked itself into a dependency on U.S. security guarantees, once again trading strategic autonomy for decades of comfortable reliance under the American umbrella.
Follow the comments:
|
