When Waymo announced its new “Waymo World Model,” it sounded almost absurd. Built on Google DeepMind’s Genie 3, the system deliberately imagines events that have never happened.
Tornadoes on highways.
Elephants crossing intersections.
Flooded streets with floating furniture.
A pedestrian dressed as a T. rex.
These aren’t glitches. They’re the training data.
For years, “AI hallucination” was treated as a flaw. But for autonomous driving, imagination is a feature. Every accident report in the self-driving industry contains some version of the same sentence: “The system had not encountered this scenario before.” World models are designed to eliminate that excuse.
From simulation to the streets
Waymo has already driven nearly 200 million fully autonomous miles on real roads, and billions more in simulation. With Genie 3, engineers can generate hyper-realistic 3D traffic environments that maintain context over time. That means consistent weather patterns, persistent objects, and believable sensor input.
The system combines 2D camera feeds with 3D LiDAR simulations to recreate exactly what the car’s 29 cameras and sensors would “see.” This creates a virtual testing ground for extreme edge cases: sudden snow in Honolulu, dense fog rolling over a bridge, or chaotic pedestrian behaviour.
It’s not just technical wizardry. It’s strategic infrastructure.
Europe: the London launch
In 2025, Waymo completed 15 million paid robotaxi rides, up from 5 million the year before. Now it’s expanding into Europe, starting with London in 2026. Unlike experimental pilots, this is a commercial rollout, fully driverless Jaguar I-PACE vehicles operating via the Waymo app.
The economic ambition is staggering. Waymo recently raised $16 billion, pushing its valuation to around $126 billion — roughly double that of Volkswagen. Analysts describe the strategy bluntly: dominate early, scale fast.
Yet profitability remains a challenge. Alphabet has reportedly invested over $35 billion into the project. Each Jaguar robotaxi costs between $150,000 and $175,000. To lower costs, Waymo is introducing the Zeekr Ojai, manufactured by Geely in China, potentially reducing per-vehicle costs by $70,000. Vehicles are imported as “gliders” and fitted with Waymo’s sensors and software in the U.S., a clever geopolitical workaround.
Safety: Promise and pressure
Safety is both Waymo’s biggest strength and biggest risk.
In Santa Monica, a Waymo vehicle struck a child who ran into the road. The system braked instantly; the company argues a human driver might have reacted more slowly. In Texas, vehicles reportedly failed to properly stop at school bus stop signs on multiple occasions. Critics warn that probabilities aren’t comfort when children are involved.
And yet public trust appears to be shifting. In San Francisco, parents increasingly use robotaxis for teenagers, seeing them as safer than unknown human drivers.
Amsterdam: A real test
Now the big question: could this work in Amsterdam?
Amsterdam’s narrow streets, dense cycling, trams, unpredictable tourists, and centuries-old infrastructure make for one of the most complex urban driving environments in the world. Add heavy rain, bridges, and delivery bikes weaving through traffic, and you have a stress test for any AI system.
Technically, the Waymo World Model is built for exactly this kind of challenge. Engineers could simulate Amsterdam’s canal streets, sudden cyclist swerves, and tram crossings billions of times before a single real-world deployment.
Economically, Amsterdam is attractive: high ride density, tourism, premium pricing potential, and strong public transport integration. However, regulatory approval, data privacy compliance, and public acceptance would be decisive factors.
The bigger picture
This isn’t about better chatbots. It’s about AI systems that simulate reality before acting in it.
If world models can truly anticipate the improbable, autonomous vehicles may finally overcome the “edge case” problem. And if that happens, cities like London, and perhaps even Amsterdam, could see transport transformed: fewer accidents, optimised traffic flow, lower emissions, and new mobility economics.
Hallucination was never the enemy.
It was just solving the wrong problem.
The future of transport may depend on machines that imagine the impossible — so they can safely navigate the real.
