Nvidia, Groq and the Trojan horse

Posted on: 21 January 2026

On Christmas Eve 2025, while Europe was raising glasses to the festive season, Nvidia announced the acquisition of Groq's assets for twenty billion dollars. The news passed almost unnoticed, buried between turkey leftovers and holiday greetings. Perfect timing, as communications professionals know well: stories you don't want debated get published when nobody's watching.

Groq was the only company in the world to have developed a credible alternative to Nvidia's graphics processors for running artificial intelligence models. Its chips, called Language Processing Units, consumed a fraction of the energy and achieved speeds up to ten times faster for inference operations: that moment when AI stops learning and starts responding, generating, working. Inference is where artificial intelligence becomes a service. It's the tap from which access to AI flows for billions of users. And now Nvidia controls both the tap and the meter.

But the real story isn't the acquisition itself. It's the structure in which it was wrapped.

On paper, Groq remains an independent company. It has a new chief executive, Simon Edwards, and formally continues to operate. But founder Jonathan Ross, along with ninety percent of the team and all the intellectual property, has moved to Nvidia through what's described as a "non-exclusive licensing agreement". An elegant formula that allows absorbing a competitor without technically acquiring it, thus avoiding the notification thresholds that would trigger antitrust scrutiny.

It's the same scheme Microsoft used with Inflection in 2024. And before that, with different variations, dozens of operations across the technology sector. The substance is always the same: buy everything that matters, leaving standing an empty shell that allows you to say "see, competition still exists".

The European Union and the United Kingdom have opened preliminary investigations. Headlines speak of "antitrust concerns" and "possible regulatory interventions". It's the script we know. We saw it with Google, fined over eleven billion euros across a decade. We saw it with Microsoft, with Apple, with Meta. Fines, indignant declarations, promises of "behavioural remedies". And then? Business continues. Companies pay, adjust some details, and the next day they're stronger than before.

Anyone familiar with Switzerland will recognise the pattern immediately.

For decades, Swiss cantons have attracted multinationals with generous tax incentives: reduced rates, temporary exemptions, preferential treatment lasting five, ten, fifteen years. Companies arrive, build prestigious headquarters, hire local staff. Politicians celebrate. Then the grace period ends, conditions change, and companies move elsewhere. What remains are half-empty buildings and memories of when "we were an attractive hub".

The mechanism with artificial intelligence is identical in structure, but infinitely deeper in its consequences.

Europe lacks the capacity to design and manufacture the chips needed to power AI. Commissioner Virkkunen admitted it explicitly: "Companies from third countries, especially from the USA, provide the necessary technology". Only four percent of global cloud infrastructure is European-owned. The data centres running artificial intelligence on the continent belong to Amazon, Microsoft, Google. Europe is a tenant negotiating with the landlord.

In this situation, antitrust isn't a tool for competition. It's a tool for extraction. Europe cannot block Nvidia without committing technological suicide. But it can raise its voice, threaten investigations, invoke the spectre of billion-euro fines. And then sit at the table to negotiate.

What does Europe want? The answer lies in official documents, for those with the patience to read them.

The Cloud and AI Development Act, expected in the first quarter of 2026, aims to triple European data centre capacity in five years. AI Gigafactories have gathered seventy-six expressions of interest for over two hundred and thirty billion euros in potential investment. The mantra is "digital sovereignty": the majority of owners must be European, but third-country companies can participate. Translated: come, build, hire our engineers, and we'll let you operate.

The European Council on Foreign Relations puts it without circumlocution: Europe could "soothe Trumpian ire by helping America keep its advantages in particular areas, such as cutting-edge chips and AI compute infrastructure". It's the language of negotiated surrender. We let you dominate the technology, you give us something in return.

And here enters the social calculation.

The European Union estimates that twelve million jobs will be eliminated or transformed by AI over the next three years. Goldman Sachs speaks of three hundred million globally, with the United States and Europe as the most affected areas. The World Economic Forum predicts eighty-five million jobs lost by 2026. These are numbers no welfare system can absorb without devastating political consequences.

Antitrust thus becomes the mechanism for extracting compensation. Not chip factories, impossible to build in useful timeframes. But local data centres creating a few thousand jobs. Reskilling programmes funded by Big Tech. Welfare contributions to cushion the social impact. Germany has already launched "AI UpSkill 2030" with four and a half billion euros. South Korea has introduced a robot tax. London has announced free AI training after the mayor spoke of "imminent mass unemployment".

The Groq deal offers Nvidia the perfect bargaining chip.

The "non-exclusive" structure allows telling Europe: we can license this technology to European partners too. We can build AI Gigafactories on your territory. We can hire your engineers, train your workers, contribute to your social programmes. In exchange, turn down the antitrust volume and let us work.

It's the Trojan horse in its most refined form.

Phase one: Nvidia invests in local infrastructure, transfers some expertise, hires Europeans. Europe celebrates the "concessions obtained".

Phase two: European AI infrastructure runs entirely on Nvidia hardware, Nvidia software, Nvidia protocols. The "local partners" are actually dependent system integrators. European engineers work on Nvidia projects, not on alternatives.

Phase three: the grace period ends. Conditions change. And at that point, what do you do? Switch everything off? Nationalise? With what competence, with what alternative?

The difference from Swiss tax incentives is that you can rebuild a watch factory. You can repurchase a data centre. But expertise in artificial intelligence is cumulative. Every passing year widens the gap. You're not renting a warehouse. You're ceding the learning curve itself.

DeepSeek, the Chinese startup, has demonstrated that competitive AI can be built with fewer resources, challenging Nvidia's "more compute equals more money" model. But DeepSeek is Chinese. And Europe cannot get into bed with Beijing to escape Washington. It's trapped in its dependency, forced to negotiate with the only supplier it has.

The contract has a clause written in small print that nobody reads: "Terms subject to unilateral modification".

When Nvidia decides the crumbs are too many, or that Europe has nothing left to offer in exchange, the rules will change. And at that point we'll discover we weren't negotiating our house rental. We were negotiating how much oxygen they let us breathe.

The horse is already inside the walls. And we're still debating whether the wood is of good quality.