After Hinton’s Verdict: A European Answer That Doesn’t Copy Silicon Valley
Yesterday Geoffrey Hinton, the man who taught machines to dream, gave his verdict: Google will win the AI race.
Not maybe. Not probably. Will.
He is almost certainly right — if we keep measuring victory the way Silicon Valley measures it: more parameters, higher benchmarks, faster releases, bigger market caps.
Google has the chips, the data oceans, the talent magnet and now the momentum. Gemini 3, Nano Banana Pro, the quiet burial of OpenAI’s short-lived lead. The scoreboard does not lie.
But Europe never signed up to play that game.
The frontier-model arms race is over for us. And that is not a tragedy; it is a liberation.
The race we lost (and why it no longer matters)
| What the winners have | What Europe will never have at the same scale |
|---|---|
| Hyperscale compute | Hyperscale compute |
| Consumer data firehoses | Consumer data firehoses |
| $100 billion war chests | $100 billion war chests |
| Willingness to move fast and break societies | (thank God) |
We spent years feeling guilty about this list.
We don’t have to anymore.
Because the next phase of artificial intelligence is no longer about who builds the smartest general model in a lab in Mountain View or Shenzhen.
It is about who gets to run intelligence inside hospitals, courts, power grids, factories and classrooms across the continent.
And in that phase, the rules change.
Phase Two: Deployment in the real world
By 2027 the majority of economic value from AI will not come from frontier research labs.
It will come from systems that are trusted enough to be allowed inside regulated, high-stakes environments.
- A French hospital will not hand its oncology data to an American closed model that can change its weights tomorrow without notice.
- A German car manufacturer will not let its production-line decisions be taken by a model whose training data location is classified.
- A Dutch court will not accept legal reasoning from a system that cannot show its sources under the AI Act.
Trust is becoming the scarcest resource in AI.
And Europe still owns most of it.
The Brussels Effect, quietly winning
The AI Act is not a brake; it is a moat. Every company that wants access to 450 million prosperous, highly regulated citizens has to play by European rules. That is not bureaucracy. That is market power.
Meanwhile, something else is happening under the radar:
- Mistral, Aleph Alpha and a dozen smaller labs are proving you can reach the global top-10 with budgets that are rounding errors for Google.
- Industrial giants — Siemens, Airbus, ASML, SAP — are pooling proprietary data with local AI teams to build domain-specific models that no generalist player can match.
- European cloud providers (OVH, Ionos, Scaleway) are growing 30–50 % year-on-year in exactly the sectors that refuse American hyperscalers.
This is the meso-layer we wrote about earlier: the quiet rewiring of entire value chains. It does not produce the sexiest demos. It produces the most defensible revenue.
Wisdom is the new compute
Hinton built the mathematics of intelligence.
Europe still remembers the philosophy of it.
We do not need to out-compute California or out-subsidise Beijing.
We need to out-trust them.
And then sell that trust — packaged as compliant, explainable, sovereign AI services — to the rest of the world that is slowly waking up to the cost of unchecked scale.
The race is not over.
It simply moved to a track where centuries of rule-of-law, individual rights and human-centred institutions are no longer historical baggage.
They are competitive advantage.
