Tesla's European fleet carrying HW4 — the fourth-generation self-driving computer, and the hardware current Full Self-Driving builds are written for — is roughly 548,000 cars. About 108,000 of them are registered in one of the five EU countries where FSD (Supervised) is legally usable. That is 19.7%. The other 440,000 have the hardware, can receive the software over the air, and are not allowed to switch it on.
Both fleet counts are claimed rather than disclosed. They come from Roland Pircher, an independent analyst who publishes weekly European Tesla registration counts country by country. Tesla does not break out its installed base by hardware generation or by market, so nobody outside the company can check 548,000 against a filing, and this article does not pretend otherwise. What can be checked is the perimeter: the five countries he names are exactly the five Tesla itself names.
That is the company's own European account on 1 September, and it supplies the one figure Tesla is willing to publish: more than 70,000 customers using FSD across those five markets.
The gap is not a hardware problem
The interesting thing about the 440,000 is that nothing about them needs to change. They are the same cars, running the same build, sold by the same company. Where they are parked is the entire difference. As of 7 September:
- Approved, five states. The Netherlands, Denmark, Belgium, Estonia and Lithuania. Together 8.9% of the EU's population — our arithmetic on Eurostat's 1 January 2025 figures, and it matches the 8.9% the independent FSD Europe Tracker computes.
- One authority did the work. The Dutch vehicle authority RDW granted the first provisional EU type approval on 10 April 2026, under the Article 39 exemption in Regulation (EU) 2018/858 — the route for a technology that does not fit the harmonised rules. RDW was explicit that the system "is not self-driving" and that the driver stays legally responsible.
- The other four recognised it. Lithuania, Estonia, Denmark and Belgium adopted the Dutch file rather than repeating it. That is why the approved list grew from one country to five in eight weeks and has not grown since.
- France has not approved it. On 2 September the transport ministry authorised on-road trials of two vehicles. Trials are not approval.
- Germany has not approved it. The federal motor authority has been reviewing the Dutch file since spring and has granted nothing.
- Twenty states have made no national decision at all — 57.3% of the EU's population, including Italy, Spain and Poland.
What actually closes the gap
Article 39 gets a system onto the roads of the member state that approves it, and of whichever neighbours choose to recognise that approval. It does not make the product European. For that, the European Commission has to adopt an implementing act extending the Dutch approval EU-wide, and an implementing act goes to a vote in the Technical Committee on Motor Vehicles.
That committee votes by the ordinary EU qualified majority, and the threshold is a double one: at least 55% of member states — 15 of 27 — representing at least 65% of the Union's population. Those two numbers are the whole story, and they are why this is arithmetic rather than atmosphere.
Run them. The EU-27 is 450.4 million people. Sixty-five per cent is 292.7 million. The five approved states are 40.2 million — the vote needs another 252.5 million on top of a base that is one seventh of the way there.
Now put the two holdouts back in. Germany is 18.6% of the EU's population and France is 15.2%; together 33.8%. Take them both out of the yes column and everyone else in the Union adds up to 66.2% of the population — which means clearing 65% without either of them requires yes votes from states holding 98.2% of everyone who is left. Without Germany alone it takes 79.8% of the remainder; without France alone, 76.7%.
France and Germany are not two more countries on a list. They are close to arithmetically mandatory. That is the mechanism the fleet numbers are pointing at, and it is why an article about 440,000 cars ends up being an article about a committee.
The same arithmetic runs the other way, which is the part worth holding onto. A blocking minority needs at least four member states holding more than 35% of the population. Germany, France, Italy, Spain and Czechia have all been reported as preferring a coordinated EU decision to a national clearance Brussels could later unpick. Those five are five states holding 60.2% of the EU — comfortably past the four-state, 35% bar, assembled out of countries that have not said no to Tesla so much as said not yet, and not alone.
What changed in France, and what did not
France is the reason this piece is being written now rather than in April, and the change is real but smaller than the coverage suggests.
In July the transport minister, Philippe Tabarot, said Paris did not intend to authorise the system, citing two specific objections: that it would exceed posted speed limits when matching surrounding traffic, and that it did not adequately guarantee driver attention in urban settings, at intersections and roundabouts. On 2 September he met Elon Musk, called the exchange constructive, and announced that two vehicles would be made available for testing: "With two vehicles being made available, we are now entering a new phase: that of on-road trials."
That is a country that said no in July running tests in September. It is not a country that has approved anything, and the distinction matters more here than usual, because everything on offer is supervised — a Level 2 assistance system with a legally responsible driver in the seat. France moving from refusal to trials changes the odds on a vote; it does not put a single French car into the approved column.
Germany is at a similar stage by a different route. Its federal motor authority has had the Dutch technical file since spring and asked for a vehicle of its own to evaluate. Berlin's stated preference — like Paris's — is a harmonised EU decision rather than a national recognition. Nothing has been granted. On the robotaxi side of the same regulatory axis, Germany's supervised pilots keep an employee in the driver's seat for legal reasons, which is the clearest available statement of where the country currently is.
The date everyone is now pointing at is 6 October 2026, when the committee is expected to take up the EU-wide approval. Treat that as expected, not scheduled: it is sourced to a German transport ministry email and to Tesla's own characterisation — the company's word for it is a "potential" vote — and an earlier committee meeting on 30 June produced discussion and no ballot. Nor is a yes a formality: the European Transport Safety Council has publicly called for the implementing act to be paused, arguing that a committee meeting in private is the wrong forum to decide whether hands-off assistance belongs on non-motorway roads.
If you want to follow the openings as they happen, fsdlifetracker.com is the site that has made the argument its subject: it scores lives it estimates FSD has saved where the software is permitted against "lives lost to regulation" where it is not. Read it as advocacy with a map, not as a status board — its country data is stamped February 2026 and still flags all five approved states as blocked. For the approval status itself, the FSD Europe Tracker is the one keeping pace.
Where permission arrived, attach went up — how far up, nobody outside Tesla knows
Here is the cross-check that makes the gap worth a valuation reader's time. Tesla says more than 70,000 customers use FSD across the five approved states. Pircher's count of eligible cars in those same five states is 108,000. If the two share a perimeter, attach there is at least 64.8%.
This is where most analysis would print 64.8% and move on. We are not going to. Tesla said "used by", not "paying for", and free trials have historically been part of how that number grows; the 108,000 denominator is claimed rather than disclosed; and one figure is a company statement while the other is an outside estimate. What survives is a bound, and the bound is still the finding: in the markets where the software is legal, adoption is running far above the 28% attach our own model assumes across Tesla's global HW4 fleet. Permission does not nudge this line. It switches it on.
What this does to our model
Our TSLA model sizes the FSD opportunity as a pool of 6 million HW4-capable vehicles — 1.7 million subscribers against it, 28% attach — and that definition is explicitly about hardware, not about law. Europe looks at first like a hole in it: 440,000 cars counted as addressable that are not currently allowed to address anything, 7.3% of the pool.
Take them out and see what happens. Cutting the pool from 6.0 million to 5.56 million moves fair value from $272.39 to $270.51 — $1.88 a share, on a model where the whole FSD line is worth $30.34, or 11.1% of the base case. Deleting every unapproved European car from the opportunity is a rounding error.
The reason is that the pool was never the slider carrying the regulatory risk, and our own model says so: the ceiling we place on how much of the fleet ever buys the software is where approvals live, not the count of eligible cars. The sensitivities agree, and the comparison is the point:
| Change to the FSD line | Fair value | vs $272.39 base |
|---|---|---|
| Remove Europe's 440,000 unapproved cars from the pool | $270.51 | −$1.88 |
| Cut the ceiling on lifetime attach from 90% to 82.7% | $270.31 | −$2.08 |
| Cut that ceiling to 80% | $269.54 | −$2.85 |
| Cut that ceiling to 70% | $266.69 | −$5.70 |
A ten-point cut in how much of the fleet we think ever buys FSD costs more than deleting every unapproved car in Europe. So the model does not change, and the size of the no is the useful part: a car in Munich that cannot legally run FSD today is not a car outside the opportunity. It is a car at zero attach, in a model that already assumes approvals arrive.
The honest reading of that is deflationary in both directions. It means the European regulatory story — the one that fills the automotive press every week — is worth under two dollars a share on our numbers. It also means that if the committee's answer is never rather than not yet, the damage does not show up as a smaller pool. It shows up as a ceiling that never gets reached, which is the slider that actually moves the stock.
What to watch
- Whether the committee votes on 6 October at all. The 30 June meeting produced discussion and no ballot. A second adjournment is itself information about the majority.
- What France does after the trials. Two cars on French roads is a test, not a position. The tell is whether Paris moves to recognise the Dutch approval nationally or keeps holding for the EU act.
- Whether Germany's motor authority issues anything in writing. Germany is 18.6% of the population and there is no realistic 65% without it.
- Whether Tesla ever discloses FSD subscribers by region. The 19.7% and the 64.8% both rest on an outside estimate of a fleet only Tesla can count. One line in a shareholder letter would settle both.
- Whether adoption in the five approved states holds above 60% once free trials lapse. That is the number that tells you whether permission is worth what the ramp assumes.
Sources and provenance. Europe's HW4 fleet counts — the total, the approved-country subset and the per-country split — are claimed figures from independent analyst Roland Pircher's 27 August post, compiled from national registration data and not checkable against any Tesla disclosure; the ratios drawn from them are ours. The five-country approval list and the customer count are Tesla's own statement of 1 September, and the attach floor derived from it is a bound rather than a measurement. The Dutch approval, the Article 39 route and the qualified-majority thresholds are as published by the Dutch vehicle authority, the Commission and the Council; the population shares computed on them are ours, on Eurostat's 1 January 2025 EU-27 figures. France's trial authorisation and the minister's quote are press-reported from 2–3 September, and the 6 October committee date is expected rather than confirmed. Fair values and every sensitivity come from our own TSLA model, as of 16 August 2026 against a $405.05 price that day.
