The Sign Error in Every Chinese OEM's Europe Plan
Building in Europe is sold as de-risking. On one line it does the opposite — and the meter that proves it reads near-zero today, climbs for a decade, and sits on no single person's dashboard.
These are my personal views, written from public sources only. Nothing here reflects the position of any employer, and no internal or confidential information is used. Where I state a fact I cite it; where I state a read, I say so.
When a Chinese carmaker chooses to build in Europe rather than ship cars in, the slide that backs the move always has a column labelled de-risking. Localise, and the tariff problem softens. Localise, and the minimum-price undertaking eases. Localise, and the political optics get better. Every line on that slide leads to the same conclusion: building here cuts your exposure.
One line is missing from that slide. On that line, the arrow goes the other way.
For two years the debate over Chinese cars in Europe has been framed as an argument about the height of a wall. First the tariff, then the minimum import price that replaced it, then the safety and cybersecurity rules. How high is the barrier, and can they get over it. That is the wrong question, and it pushes companies to optimise the wrong variable.
Europe does not price the Chinese auto industry by the height of its barriers. It prices it by what you have to become to operate there continuously. And the costs that settle that question are not the ones on the entry slide. They are smaller, they keep running, no single person owns them, and at least one of them moves the opposite way from what the slide assumes.
The thing that catches a second-tier entrant is not the wall it can see. It is the meter it mispriced, because nobody was assigned to read it.
Two columns, and a missing one
Put the cost of Europe into two columns.
The first column is tolls. Type approval and homologation, paid once per vehicle type. The GSR2 safety kit, a bill of materials fixed per platform. These look like entry costs, they are big, and they end. You pay the toll and the gate goes up. They are also owned: someone’s job is the homologation programme, someone signs off the safety kit. A cost with a name beside it rarely catches you cold.
The second column is meters: the costs that never turn off. UN R155 and R156, the cybersecurity and software-update management systems, re-audited every three years for the life of the platform. NIS2, the corporate-security obligation, ongoing risk management and incident reporting for as long as you operate. The Battery Regulation’s digital passport, per cell, mandatory from 18 February 2027. The Data Act’s machinery for sharing vehicle-generated data, standing infrastructure you build and keep running. None is a one-off payment. Each keeps running for as long as the business does.
The entry business case models the first column carefully. It models the second loosely, if it models it at all. But the line that actually changes the decision sits in neither column. It only shows up when you move from importing finished cars to building them in Europe — and to see it you have to do something nobody on the entry team is set up to do, which is net three effects that sit in three different departments.
So — if the only European numbers you can name are the tariff and the homologation, you have priced the entry and ignored the operation.
First, the honest ranking
Before I say anything about the meters, it is worth being plain about where they rank, because the temptation in a piece like this is to dress up a clever small cost as the deciding one. It is not.
Rank a Chinese entrant’s annual cash burn in Europe and the regulatory meters do not sit at the top. The service network burns more: sites, high-voltage technicians, parts logistics. Brand and marketing burn more. Residual-value support burns more, and it is burning now — German valuation group DAT shows Chinese EVs depreciating at roughly twice the market average, and leasing firms like Arval have raised prices, with some demanding compensation upfront before they will take the cars at all. In a market where up to eighty per cent of EVs are leased, that is a direct, scaling drain.
You do not need to theorise about which costs bite hardest, because one entrant is showing it to you in real time. NIO registered eight cars in Germany in the first quarter of 2026 — one in January, five in February, two in March — and is seeking subtenants for its four flagship showrooms in Berlin, Frankfurt, Düsseldorf and Hamburg, dismantling its European management and moving to a distributor model. It presents this as a reset, not an exit, and says it is staying in Europe. But what it was forced off was its cost structure: company-owned showrooms, service hubs and swap stations, unsustainable at modest volume. NIO was not forced off its model by a compliance meter. It was forced off by the most visible, most owned, most budgeted costs on the slide.
So I am not going to tell you the meters are the biggest cost, or that they decide survival. They are not. They are the cost that is both not the biggest and not owned — small enough to miss in the entry case, spread across enough functions that no one totals them, and, on one axis, pointing the opposite way from what the entry team assumes.
So — the danger is not that the meters are the largest cost. It is that they are the largest unexamined one.
The lever that is actually a knob
Here is the line that is missing from the slide, and the real size of it.
The entry team treats localisation as one lever: build in Europe, and the tariff exposure falls. It is not one lever. It is a single knob — how much of the car you actually make in Europe — and that knob sets three things at once, two of them working against each other.
Turn it low — shallow assembly, a kit shipped in and bolted together — and you do not get the tariff relief you came for. The EU’s anti-circumvention test treats local assembly as circumvention, and keeps the anti-subsidy duty on the EU-built car, when the parts shipped in are 60% or more of the total parts value and the value added in Europe is 25% or less of the manufacturing cost; completely-knocked-down kits are watched for exactly this. Shallow assembly fails that test. You built a plant and still pay the duty.
Turn it high — real manufacturing, enough local value to clear the threshold — and the tariff relief is real. But now you are importing steel and aluminium as raw materials to feed an actual line, and that is what trips the line on no one’s slide: CBAM, the EU’s carbon border levy.
CBAM is widely assumed to tax the carbon in whatever you import, including a finished car. It does not. It applies to six upstream materials at the border, by customs code: iron and steel, aluminium, cement, fertilisers, hydrogen, electricity. A steel coil is in scope; a car body is not. A vehicle imported whole from China carries no CBAM at all. So the exposure runs backwards to the slide: importing finished cars is the CBAM-free path, and the deep-manufacturing setting that earns the tariff relief is the one that creates the carbon exposure.
Two conditions keep this straight. It only fires if those raw materials are imported from outside the EU — buy your steel in Europe and it sits under the ETS, not CBAM, and a carmaker selling a localisation story has every incentive to source locally, which can switch the exposure off. And the magnitude is back-loaded by design. CBAM only charges the slice of embedded carbon that EU free allocation no longer covers, and that slice phases in: 2.5% in 2026, 10% by 2028, about half by 2030, 100% by 2034.
Put real numbers on it. A car carries on the order of 900 kg of steel and about 150 kg of aluminium; at BF-BOF steel near 2.1 tonnes of CO₂ per tonne and high-carbon non-EU aluminium far higher, the embedded carbon in those imported materials runs a few tonnes. At a CBAM price around €75 a tonne, the fully phased levy is on the order of €200–350 a car. [BET — assumption stack; inputs sourced, total mine] But fully phased is 2034. In 2026 the 2.5% factor makes it roughly €6–10 a car — trivial, the same order as the run-cost meters I just told you not to over-rate. By the early 2030s it is in the low hundreds.
So the honest answer is not “localising arms a big hidden cost.” It is sharper and more awkward than that: localising, if you manufacture deeply enough to earn the tariff break and source your raw materials outside the EU, installs a meter that reads near-zero today and climbs for a decade. The entry spreadsheet, which prices the 2026 tariff saving, is the wrong horizon. A plant is a fifteen-year asset. The number that belongs in that decision is the 2032 one, and almost nobody is putting it there.
So — the localisation decision is not a de-risking. It is a re-pricing across three variables and a decade, and the one knob that drives them is on no single person’s dashboard.
Why nobody nets it
The knob has no owner because its three effects are three different kinds of rule, filed in three different departments.
Sort the European stack by who owns it and it splits five ways. Type approval, the GSR2 kit and the vehicle side of R155 sit with product and engineering. The cybersecurity management system and NIS2 sit with corporate IT. The tariff, the minimum import price and CBAM sit with trade and finance — but the local-content level that drives all three is set by manufacturing strategy, a fourth desk. The Data Act and GDPR sit with data and legal. Five functions, and no one whose job is the cross-cutting whole.
So when manufacturing turns the localisation knob up to clear the anti-circumvention threshold, trade books a tariff win, no one goes back to check the carbon desk, and the back-loaded CBAM meter starts its decade-long climb unattributed. In a European entity that is often five to fifty people, with one to three sent from China and a handful hired locally, the person who would have netted the knob is simply not there. The trade-off is real, it has a sign and a time-curve, and it is owned by nobody.
This is the biggest unstaffed role in a Chinese carmaker’s Europe. Not a country manager. The person who can see all the meters at once and net the trade-offs between them. Most entrants have not hired that person, and the org chart they brought from home has no box for them.
So — the cost that surprises you is not the one you forgot to budget. It is the one no one was assigned to total.
The asymmetry is in the people, not the per-car euros
There is an asymmetry in the run-cost meters as well, and it is better to state it cleanly than reach for some dramatic per-car figure.
The arithmetic, honestly done, strips the drama out. Take NIS2 on its own: Germany’s own legislative cost estimate is about €70,000 to set up and €30,000 a year to run, per entity. That is one meter among several. Add the fixed pieces — a CSMS audit and its three-year re-audit, the NIS2 function, battery-passport integration, the data build — and the annual fixed regulatory run-cost ends up in the low single-digit millions. [BET — assumption stack, components sourced, total not] Spread over a hundred thousand cars, that is tens of euros a car on a thirty-five-thousand-euro car. As a per-vehicle figure it is trivial, and any honest account of this needs to say that plainly.
The asymmetry is not in the euros per car. It sits in the people. Volkswagen runs a CSMS, a NIS2 function, a battery-compliance team and a data-governance office as departments it built over decades and spreads across millions of cars. A fifteen-person European entity meets the same obligations with the same fifteen people. The smallest entrant carries the same org chart of obligations as the largest, with only a fraction of the headcount to throw at it. That is the regressive cost — not a line item, a staffing floor that does not scale down.
So — the meter you cannot amortise is not measured in euros per car. It is measured in functions per head.
The strongest version of the other side
The strongest case against all of this is a serious one, and I have to treat it that way.
Every global carmaker carries a stack like this. Volkswagen in China faces a cybersecurity law, data-localisation rules, the GB standards, the dual-credit system. This is what it costs to be a global car company, the majors carry it with departments built for the job, and the Chinese entrants will build the same. There is nothing in this a competent company does not already understand.
I concede that frame. This is the cost of being global now. But the symmetry breaks in two places. The major built these functions over decades and has someone who nets them; the entrant is building them cold, in two years, with no one in the netting seat — which is exactly how the localisation knob gets turned with only one of its three effects on the table. And the major spreads the staffing floor across millions of units; the entrant cannot.
One guardrail, because this is where it is easy to get sloppy. None of this is a machine the EU built to keep Chinese carmakers out. These rules were written at different times, by different parts of the apparatus, for different reasons. There is no systemic architect on the European side any more than there is a single owner on the company side. Coordination exists, but it is sectoral, never whole: a UNECE working group harmonises the vehicle rules, one directorate runs the carbon border, another the trade defence, each minding its own sector and none owning the cross-cutting cost. The one place the EU deliberately de-conflicted shows the shape of it — vehicles are carved out of the Cyber Resilience Act because type approval and R155 already cover them. They de-conflict two instruments when one product would be hit twice. They do not net the cost and timing of the whole. It is a thing no one owns, on either side — which is why the entrant that also leaves it unowned meets it as a surprise rather than a plan.
What changes, and for whom
If you run market entry for a Chinese OEM: build the one model nobody is building — local-content depth against all three of its effects. Below the anti-circumvention threshold you keep the tariff and waste the plant; above it you earn the tariff and switch on a carbon meter that is near-zero in 2026 and in the low hundreds per car by the early 2030s, if your raw materials come from outside the EU. The lever you think moves only the tariff sets your duty status, your carbon exposure and your sourcing strategy in one move. Price the 2032 version, not the 2026 one, because the plant outlives the phase-in. And source steel and aluminium inside the EU unless you have a specific reason not to — that single choice turns the carbon meter off.
If you are the CFO or the strategist: watch the 2028 CBAM downstream proposal, which would extend the levy to roughly 180 finished and semi-finished products including vehicle components, pending Parliament and Council. If it passes, the precursor exposure stops being only raw steel and starts reaching into the bought-in parts a European line runs on — and the localisation maths moves again, against the deep-manufacturing case you may be about to sign.
If you are a Tier-1 supplier: you already sit inside several of these meters — the R155 supply chain, the Battery Regulation’s due diligence, and, if the downstream extension passes, CBAM on the components you ship. Expect the carmakers to push that metering onto you by contract.
If you are an investor sizing one of these companies: do not ask whether they cleared the tariff. Ask who owns the localisation knob and whether they have netted its three effects. A team that can answer has seen the line that is not on the slide. A team that cannot has a sign error somewhere in its European plan, and a clock already running on it.
What I’m sure of, and what I’m betting
Keep the two separate, because that difference is the point.
What is fact, and cited above: CBAM covers six upstream materials and not finished cars; its levy phases in from 2.5% in 2026 to 100% in 2034; the anti-circumvention duty stays on cars assembled from 60%-or-more imported parts with 25%-or-less local value added; NIS2 costs a German entity roughly €30,000 a year to run; NIO is moving off direct sales on cost structure. What is my bet [BET]: that the cost which catches out a second-tier entrant is not the biggest one but the un-owned one — and that the localisation knob, which sets duty, carbon and sourcing at once across a decade and sits on no single dashboard, is the cleanest example of a whole class of trade-offs nobody is staffed to net. That is a judgment, not a law, and you should hold it that way.
The thing I am most sure of is the smallest one. The barrier you can see is priced by someone. The meter you install by accident, that reads near-zero in the year you install it, is priced by no one — which is exactly why it is still spinning when the surprise arrives. Getting into Europe was always the answerable question. Operating there without a person who nets the whole is where the quiet money goes.
You think localising moves one lever, the tariff. It is a knob that turns three things at once — and no one wired it to a gauge.
If you work on China–Europe automotive strategy and want the next one in your inbox, subscribe. I publish roughly once a fortnight, and I read every reply.