The hardware travels. The permission doesn't.

A BYD Yuan Up costs ¥89,800 in China — about $12,600 — and for that money it will follow a navigation route down a motorway on its own, park itself without you in it, and find you again afterwards. The same car is sold in Britain as the Atto 2, from £30,850, about $39,600. It will hold a lane and keep a distance. It will not follow a route. There is no version of it, at any price, that will.

That is a 3.1× price and a subtraction, and it is not a one-off. Over the past month this site has reviewed six cars where the same thing happens, and the pattern is consistent enough to deserve a proper examination rather than another paragraph tacked onto a review.

So: what actually explains it? We took the four explanations that get offered, and looked for the case that breaks each one.

First, the pattern — in the makers' own words

The evidence is not subtle, and in one case the manufacturer numbers it for you.

The BYD Sealion 7 ships in Europe with a system BYD calls DiPilot 10. In China the same car carries DiPilot 100. You do not need a teardown to read that; the product name states the gap.

The Atto 3 exports with a bought-in stack BYD declines to name, no God's Eye and no navigation assist, at roughly three times its Chinese price. The Denza D9 carries God's Eye 5.0 as standard on every Chinese variant — motorway and urban navigation, full-scene parking, automatic braking to 120 km/h, with a roof lidar even on the entry car — while the Australian version, at A$85,990 to A$95,990, gets plain Level 2 with no lidar and no navigation assist at any trim. The Zeekr 9X puts five lidar and 1,400 TOPS of Nvidia compute on its top trim in China; the dual-chip arrangement is reported as China-only. The Xiaomi YU7 fits lidar, 4D radar, eleven cameras and 700 TOPS to every single trim, and arrives in Germany in 2027 with the question still open.

One car in the set breaks the pattern and is worth naming, because it shows we are not simply pattern-matching: the BYD Shark 6 also exports with plain Level 2 — but BYD does not sell it in China at all. There is no better domestic version being withheld. Its modest driver assistance is just its driver assistance.

Explanation one: tariffs

The obvious answer in Europe is trade policy, and the policy is real. Since 30 October 2024 the EU has charged BYD a 17 per cent countervailing duty on battery-electric cars from China, for five years, on top of the standard 10 per cent import duty. Add shipping, type approval, VAT and a dealer network built from nothing, and the price gap stops being mysterious.

Except the missing thing is not the price. Software is not in the container and pays no duty at the border. And we know what it is worth, because BYD publishes it: inside China, the step from no God's Eye to God's Eye on the Yuan Up is ¥8,000, about $1,120, on identical hardware and an identical battery.

Then there is Australia, which is the control experiment this argument needed. Australia applies no anti-subsidy duty to Chinese electric cars; they arrive under a free trade agreement. The Atto 2 accordingly sells there for A$31,990 before on-roads, around $20,800 — a third less than the British car. The tariff is absent and the price reflects it. The feature is still missing. Same with the Denza D9. Two completely different tax regimes, one identical subtraction.

And if anyone doubts that BYD responds to tariffs, consider what it did about this one. Plug-in hybrids from China pay only the standard 10 per cent, so BYD launched an Atto 2 DM-i at £26,995 — below its own electric car — and by May 2026 roughly 70 per cent of its German registrations were hybrids. In June 2026 the European Commission opened an anti-subsidy investigation into Chinese PHEVs to close the gap. Faced with a tariff, BYD re-engineered an entire European range around it inside eighteen months. Faced with the software question, it has shipped nothing across four launches.

Verdict: explains the price, explains none of the subtraction.

Explanation two: the hardware is too expensive to export

This one dies on contact with the cars themselves, because the hardware goes anyway.

The Zeekr 9X is sold as “L3-ready” — hardware specified for eyes-off driving, in a car certified for hands-on Level 2. The Xiaomi YU7 fits a 200-metre lidar to its cheapest trim and drives hands-on. And the clearest case is not Chinese at all: the Volvo EX90 has a lidar on its roof that has never been switched on for the features it was fitted for. Buyers in all three cases have already paid for sensing they cannot use.

Verdict: falsified. The expensive part ships; the cheap part doesn't.

Explanation three: they reserve it for premium buyers

Also falsified, and by BYD more comprehensively than by anyone else. In February 2025 it added smart driving to 21 models in China at no extra cost. In May 2026 it put lidar on a Seagull costing about $13,000. When it cut ¥30,000 from the Denza D9 in July 2026, the money came out of interior trim and chassis components — not the sensor suite.

A company protecting a premium feature does not fit it to its cheapest car and then take the cost out of the seats.

Verdict: falsified at home. Whatever is happening at the border is not market segmentation.

Explanation four: the map

Here the story stops being about willingness and starts being about engineering, and one company has shown us exactly what the dependency is.

Navigation-assisted driving is not lane-keeping with ambition. It needs to know that this junction is an exit, that this ramp curves right at 60 km/h, that the lane you are in ends in 400 metres. In China those systems lean on domestic high-definition mapping that has no equivalent abroad, and cannot simply be carried over.

On 16 July 2026, Xpeng announced it would build its navigation on the Google Maps Auto SDK — the first Chinese carmaker to ship with it — specifically because, in its own framing, the overseas application of its assisted-driving system must rely on map data and navigation manoeuvres it does not otherwise have. Xpeng had also just completed European localisation testing in Germany, with a vice-president reporting the system handled local urban scenarios “with almost no additional data training”.

That is a real, unglamorous, expensive dependency, and it is the most honest part of the whole story. It is also, on Xpeng's own timeline, a solved problem: rollout targeted for 2027.

Verdict: genuine, and being dismantled in public.

Explanation five: Europe won't allow it — and the company that proved otherwise

The last defence, and the one we have repeated ourselves, is regulatory: that hands-off driving is simply not permitted, so there is nothing to ship.

That was never quite true. UN Regulation No. 171, the DCAS framework, makes hands-off assisted driving legal in Europe without the manufacturer taking on the liability that Level 3 demands. The BMW iX3 is certified under it and drives hands-off to 130 km/h for about €1,400.

What we wrote — in two reviews, as recently as this month — was that no Chinese manufacturer had used that route. That was wrong, and we are correcting it here and in those reviews.

On 18 March 2026, Geely's Qianli Haohan G-ASD became the first Chinese driver-assistance system certified under UN R171, in a joint approval issued by CATARC and IDIADA. It permits assisted driving at up to 150 km/h on European motorways, with automated interchanges, on- and off-ramp handling, autonomous overtaking and dynamic speed regulation, on dual Nvidia Orin-X compute. Geely and CATARC built a testing framework to cover a gap in the standard itself.

And critically, the certification works as a passport: a car carrying G-ASD can be sold across every UNECE member state without country-by-country approval. Geely says it will roll the system out across Geely, Zeekr, Lynk & Co and Lotus, with the first equipped car expected on European roads in June 2026. We have not been able to confirm that a customer car has actually shipped with it, and that distinction matters — a certificate is not a delivery.

Verdict: the barrier is real, surmountable, and has been surmounted.

What that leaves

Strip out what the evidence kills and the residue is small, specific and uncomfortable.

Tariffs explain the money and nothing else. Hardware cost explains nothing, because the hardware goes regardless. Segmentation is contradicted by BYD's own behaviour at home. Mapping is a genuine dependency with a known price and a public solution. Regulation is a genuine barrier that a Chinese group cleared eighteen months after the framework existed.

What remains is certification effort, per-market validation, and appetite for liability in courts you have never been tested in — all of which are costs, and all of which are therefore choices about priority. That is a different sentence from “they are not allowed to”, and it is the one buyers should be putting to salespeople.

We should be equally plain about the limits of this. No manufacturer has told us why. Nobody publishes a document saying “we decided the European certification spend was not worth it this year”. What a desk review can do is take each offered explanation and look for the market, the price list or the competitor that removes it — which is what we have done here.

And it is not only a Chinese story

The most useful corrective in this whole investigation is a German one.

BMW's own UK configurator states that Highway Assistant — the hands-off function on the iX3 — is not available in the UK or Ireland. A British buyer paying from £58,755 receives the heated camera, the surround cameras, the HD map and the compute, and does not receive the function they enable. The reason is type approval, not commerce. A German car's autonomy stops at the Channel.

Put that next to the Volvo EX90's dormant lidar — a Swedish-designed, American-built car from a Geely-owned brand — and the frame corrects itself. This is not a story about Chinese manufacturers short-changing Western customers. It is a story about a fact of the industry that nobody advertises:

Autonomy is a permission, granted market by market, and the hardware now travels considerably faster than the permission does.

Four things we are watching, all with dates

• Szeged, Q4 2026. BYD's Hungarian plant, 200,000 cars a year, with the Atto 2 among the announced models. Series production has already slipped from end-2025 to Q2 2026 to Q4. A Hungarian-built Atto 2 pays no countervailing duty — and if the software is still missing then, the tariff explanation is dead for good.
• A G-ASD car in a European customer's hands. Certified in March, expected on European roads in June 2026. If Zeekr or Lynk & Co is now selling motorway navigation assist in Europe, that is the single most important fact in this market and it has gone almost unreported.
• Xiaomi in Germany, 2027. Eight dealer groups signed, a car that already carries lidar on every trim, and a certification route that demonstrably works. If HAD does not cross the border, it will not be for want of a mechanism.
• Xpeng's 2027 rollout. Its chief executive has said he hopes to be first; Geely has already been first at certification. Watch whether Xpeng's Google Maps route or Geely's certification route becomes the template everyone copies.

What to do if you are buying one

One practical rule comes out of all of this, and it applies to any car sold in more than one country.

Ask what this car does in your market, on your roads, today — not what the model does. The badge, the sensor list and the press release describe a global product. The function is local, it is decided by a certificate you will never see, and on current evidence it is the single largest undisclosed difference between the car in the review and the car on your driveway.