Volkswagen's Sovereignty Trap

Volkswagen's reported failure with its Automated Driving Alliance, described by Alex Voigt, is not just another software setback.

It exposes the strategic dilemma of European industry in one case. It is the same pattern I wrote about in No Leverage, No Sovereignty and Europe Is Not Sovereign. It Is Comfortable.: Europe wants to speak the language of sovereignty without owning enough of the stack.

For years, VW tried to solve autonomy in the old European industrial way: a large alliance, a supplier ecosystem, internal software platforms, long timelines, and the belief that enough engineering coordination would eventually close the gap.

But autonomous driving is not a normal automotive subsystem.

It is not a gearbox. It is not a battery pack. It is not even software in the old sense.

It is a learning system.

And learning systems compound.

The moat is not the software

The most dangerous misunderstanding in the Volkswagen debate is that the company needs to find a better autonomy supplier.

It does not.

Volkswagen needs an autonomy learning loop.

That loop is simple to describe and brutal to build: millions of cars on the road, cameras collecting edge cases every day, data flowing back to headquarters, models improving, over-the-air deployment, and then another round of better data.

Tesla understood this early. Every Tesla on the road is not only a product. It is also a sensor in the learning machine. That is why the company's advantage is not merely that FSD Supervised is impressive. The advantage is that the fleet teaches the system every day.

XPeng increasingly understands this as well. The Chinese players are not simply copying Tesla's interface. They are copying the deeper logic: put cameras into the fleet, collect data aggressively, train models, deploy, repeat.

Most European OEMs still behave as if autonomy is something that can be sourced once the technology matures.

That is the mistake.

You can buy sensors. You can buy chips. You can buy an ADAS stack. You can license a certified highway feature. You can partner with Mobileye, Nvidia, Bosch, or another supplier.

But you cannot quickly buy ten years of accumulated reality.

The market is moving faster than the language

Europe still talks about autonomy in regulatory categories: L2, L2+, L3, L4, driver assistance, hands-off, eyes-on, certified highway systems.

That language matters legally. It does not fully describe what consumers will feel when the category tips.

In April 2026, the Dutch vehicle authority RDW approved Tesla FSD Supervised in the Netherlands under UN Regulation 171. It remains a supervised driver assistance system. RDW was clear that the car is not autonomous and that the driver remains responsible.

That distinction matters.

But so does the direction.

Once supervised FSD is allowed in parts of Europe, the next psychological step is obvious. People will see cars handling normal driving tasks with increasing competence. They will experience the distance between a traditional ADAS feature and a car that appears to drive itself through real traffic.

Then unsupervised FSD becomes the real threshold.

When that arrives at scale, autonomy stops being a luxury feature. It becomes the reason to buy the car.

That is the Kodak moment for legacy OEMs.

Not because their current cars are bad. Not because Mercedes, BMW, or VW cannot build beautiful machines. They can.

But because consumers may suddenly understand that a car without unsupervised autonomy is yesterday's hardware with a battery.

Cybercab changes the imagination

The Cybercab matters because it makes the abstraction physical.

Tesla is not only talking about autonomous software. It is building a dedicated vehicle around the assumption that the driver disappears. No steering wheel. No pedals. A simplified design. A manufacturing system designed for very high volume, with Cybercab production targets already being framed around a different manufacturing model.

Even if timelines slip, the direction is what matters.

If Tesla ramps Cybercab production seriously, and if Berlin becomes part of that European production path, the taxi market is the first visible target. A purpose-built autonomous electric taxi running at high utilization attacks the economics of urban mobility directly.

But the deeper disruption is not taxis.

The deeper disruption is consumer expectation.

Once people see unsupervised vehicles operating in real life, every other EV will be judged against that possibility. Why buy a car that cannot drive itself if another one can? Why pay premium prices for a vehicle whose intelligence does not compound?

This is how categories break. Slowly in technology, then suddenly in perception.

Volkswagen's real options

Volkswagen now has three real routes.

The first is the American route: make a serious long term deal with Tesla.

This is politically painful. Europe does not want to depend on a US company. It especially does not want to depend on Tesla. And certainly not on Musk, in the current political and cultural climate.

But strategy is not therapy.

If survival is the criterion, Tesla is the most rational partner because Tesla has the strongest fleet learning loop in consumer autonomy. It has the cars, the cameras, the model training, the OTA system, the internal hardware-software integration, and the cultural tolerance for shipping, learning, and improving in public.

A Volkswagen-Tesla deal would be humiliating for Wolfsburg and uncomfortable for Brussel.

That does not make it wrong.

The second route is China.

Volkswagen could avoid Tesla and choose a Chinese autonomy stack instead: XPeng, Huawei, Baidu, or another player in that ecosystem. Technologically, this route is becoming more plausible every year. Chinese companies understand the camera-data-learning loop much better than most legacy OEMs.

Politically, this may even be easier to sell in some European rooms because it avoids the emotional charge around Musk and the current United States.

But for European sovereignty, this is not obviously better.

Europe would avoid dependence on Tesla only to become dependent on China for the intelligence layer of mobility.

That is a strange victory.

The third route exists, but barely.

Volkswagen could attempt a European autonomy moonshot: go all-in on Mobileye as the automotive autonomy base, connect it to European compute and AI capacity through players such as ASML and Mistral, and turn every VW EV sold from now on into part of a European learning system.

But this route only starts with one non-negotiable move.

Put eight cameras in every EV today.

Not as an option. Not as a premium package. Not as a feature consumers must pay for. Not after the software is ready.

Now.

Because the cameras are not the feature. They are the sensor network for the intelligence that has to be built.

If Volkswagen is not doing that, the European route is not a strategy. It is window dressing. Another expensive setup for failure.

Sovereignty is not pretending you are early

Europe keeps confusing sovereignty with independence.

Real sovereignty is not the fantasy that you can build everything yourself after losing the compounding race. Real sovereignty is the ability to choose your dependencies while you still have leverage.

That is the painful part for Volkswagen.

If it chooses Tesla, it admits that the most important software layer in the future car was built outside Europe.

If it chooses China, it may preserve some political distance from Tesla while walking into a deeper strategic dependency.

If it chooses the European moonshot, it has to stop behaving like an OEM buying components and start behaving like a machine learning company with factories attached.

That means cameras everywhere. Data collection everywhere. Model improvement as the core industrial process.

Not someday.

Today.

The old way will not save VW

A supplier stack can give Volkswagen features.

It cannot give Volkswagen the compounding knowledge of a fleet that has been learning for years.

This is why another internal mega-project would be so dangerous. It would feel responsible. It would create roadmaps, committees, budgets, architecture diagrams, compliance frameworks, and European industrial language.

It would also lose more time.

The brutal question is not whether VW can build a good driver assistance system. It probably can.

The question is whether VW can build, own, and operate the learning loop fast enough before unsupervised autonomy becomes the buying criterion.

If the answer is no, then the choice is already narrowing.

US intelligence layer.

Chinese intelligence layer.

Or a European moonshot that only deserves the name if the cameras are already going into every car.

Everything else is theatre.

Volkswagen can still survive this transition, but only if it stops protecting the illusion of independence.

A Tesla deal would be humiliating for Europe. A Chinese rescue would be worse. A European autonomy stack would be beautiful, but only if Europe finally does the operational thing it has avoided for years: collect reality at scale.

The choice is not between sovereignty and dependency.

The choice is between chosen dependency now, forced dependency later, or the rare courage to build the learning machine before the market closes.

Later will be much more expensive.