Clean on Paper
The Bill Behind the EV Transition
A recent interview with a Slovenian economist on the future of electric mobility was sharp on the industrial economics — Chinese battery dominance, European manufacturers caught flat-footed, the gap between regulatory ambition and production reality. It was also, like most mainstream coverage of the EV transition, silent on the deeper structural problems. The romantic phase he identified is real. What follows the romance is what this essay is about.
I. The Battery Dependency
The standard account of Chinese EV dominance focuses on manufacturing cost. Chinese factories are more automated, labour is cheaper, scale is larger. All true, but it misses where the advantage is actually rooted.
China’s position in the battery supply chain is not primarily about assembly. It is about control of the processing layer — the industrial chemistry that sits between raw materials in the ground and finished cells in a vehicle. This layer took roughly fifteen years of deliberate state-directed industrial policy to build: protected domestic markets during the learning curve, patient capital from state banks, guaranteed demand from domestic automakers contractually required to source locally. The result is that CATL (Contemporary Amperex Technology, China’s dominant battery manufacturer) and BYD (China’s largest electric vehicle maker) together control roughly two thirds of global battery cell production, and Chinese firms control the majority of the processing capacity that feeds them.
Europe’s response was to try to build one node of this chain — the cell factory1 — without the others.2 Northvolt, the Swedish startup that attracted billions in subsidies and investment from Volkswagen, Goldman Sachs, and the EU, collapsed into bankruptcy in 2024. The diagnosis is instructive: Northvolt had production capacity but no guaranteed offtake. European automakers continued sourcing cells from CATL rather than committing to the domestic supplier that needed their volume to achieve viable production economics. Without long-term purchase contracts, the factory could not justify its capital costs. The same failure mode is visible today in European green hydrogen: roughly 90% of producers have no confirmed customers, which means no project can achieve the scale needed to reduce costs, which means no customers materialise.
You cannot build one link of a supply chain in isolation. China understood this. Europe, committed to market mechanisms at precisely the moment when industrial policy was needed, did not.
II. The Materials Dependency
Beneath the battery dependency lies a materials dependency that is more geopolitically complex and less discussed.
Lithium, the foundational element of every major battery chemistry, is extracted primarily in Chile, Argentina, and Bolivia — the lithium triangle — as well as Australia, whose large hard-rock deposits make it a major producer outside the triangle. But extraction is only half the story. China controls roughly 60-70% of global lithium refining capacity, including material it does not mine. The processing bottleneck is as strategically significant as the ore deposit, and Europe has negligible refining infrastructure for any battery material.
Cobalt is the sharper case. Approximately 70% of global cobalt supply comes from the Democratic Republic of Congo (DRC), a state whose governance structures make reliable supply chain auditing close to impossible. Chinese companies dominate both extraction and processing. The human rights documentation around artisanal cobalt mining in the DRC, including well-evidenced child labour in small-scale operations, has created pressure on battery manufacturers that the industry has partly resolved by engineering cobalt out of cheaper chemistries. BYD’s preferred lithium iron phosphate (LFP) cells — a cheaper battery chemistry containing no cobalt — are a meaningful part of why BYD’s cost structure is structurally different from manufacturers still on cobalt-containing chemistries. The tradeoff is energy density: LFP stores less energy per kilogram, which limits range and adds weight, making it more suitable for urban and budget vehicles than long-range premium models.
Two further chokepoints are almost entirely absent from public discourse. Rare earth elements — neodymium, dysprosium, praseodymium — are not used in batteries at all but in the permanent magnets of electric motors. China controls roughly 85-90% of both mining and processing of these materials. This is a separate dependency from the battery chain but equally critical to EV production, and conflating the two (as coverage routinely does) obscures both. Then there is graphite, used in battery anodes: China processes roughly 95% of the world’s spherical graphite.3 Almost never mentioned. Entirely Chinese-controlled.
The EU Critical Raw Materials Act, passed in 2024, correctly diagnoses these dependencies. Its implementation timelines are optimistic and its processing gap provisions are weak. Naming a problem is not solving it.
The honest comparison with oil dependency is uncomfortable. European reliance on Middle Eastern and Russian hydrocarbons was at least a dependency on transactional states whose behaviour was legible and whose leverage was understood. The new dependencies are on a more complex map: a Chinese processing monopoly built through patient industrial policy and not easily replicated; a Congolese extraction sector embedded in a failed state; resource nationalism currents in Latin American democracies that have already produced one attempted nationalisation of lithium reserves. Whether this constellation is more or less stable than the old one is genuinely uncertain. That nobody is asking the question is not.
III. The Grid Nobody Upgraded
The electricity that will power Europe’s EV fleet has to come from somewhere and travel through something. Both the somewhere and the something are under-examined.
The grid problem has two layers. The first is load profile. Full electrification of Europe’s vehicle fleet adds roughly 15-20% to total electricity consumption — manageable in aggregate over two decades if investment proceeds steadily. What is less manageable is timing. EVs charged predominantly at home in the evening create a demand spike that coincides with existing peak household consumption. The binding constraint is not total generation capacity but local distribution infrastructure: the neighbourhood transformer, the low-voltage cable, the substation that was sized for a street of houses and is now expected to charge twenty vehicles simultaneously. Upgrading this layer requires physical crews, planning permissions, and years of sequential work. It cannot be accelerated by policy announcements.
The second layer arrived faster than anyone planned for. Hyperscale AI data centres draw 100-500 megawatts continuously, 24 hours a day, with no tolerance for intermittency. A single large facility consumes as much electricity as a medium-sized city, and several are being approved and built within 18-24 months of planning consent. Ireland’s grid operator has imposed an effective moratorium on new data centre connections in the Dublin area. The Netherlands has faced similar pressure. These are not edge cases — they are early signals of a demand shock that is accelerating as AI infrastructure investment intensifies globally.
The two shocks — EV fleet electrification and AI compute — are treated as separate stories in separate literatures. They compete for the same infrastructure response: grid investment, new generation capacity, distribution network reinforcement, and the physical supply chain that enables all of it. Transformer delivery times across Europe currently run to 18-36 months4 because every grid operator on the continent is upgrading simultaneously. Copper supply is tight. Engineering capacity is constrained. And a significant share of the equipment needed to build the grid that will reduce European energy dependency is manufactured in China — transformers, inverters, grid-scale battery storage systems. The dependency is being infrastructurally embedded at the moment of attempted escape.5
IV. The Generation Question
The electricity powering the transition has to come from somewhere, and the honest answer to where is less comfortable than the policy documents suggest.
The renewables-only framing assumed that wind and solar, scaled sufficiently and connected across borders, could carry the continental load. The AI data centre demand shock makes this untenable. A facility running tens of thousands of GPUs cannot shift its load to when the wind blows. The baseload requirement is absolute, and no amount of interconnection or demand flexibility resolves it. The corporate sector has drawn its own conclusion: Microsoft contracted to restart Three Mile Island in Pennsylvania specifically to power data centre expansion, Google and Amazon have signed agreements with multiple nuclear developers. These are engineering decisions, not ideology.
In Europe the implication is equally clear even if the politics are not. Existing nuclear is the most undervalued asset the energy transition possesses. France, which maintained its fleet, is a net electricity exporter. Germany, which completed its exit in April 2023, is burning more gas than it would otherwise be — a carbon regression justified by arguments that have not aged well. Extending the operational life of functioning plants is the single highest-return decision available to any European government that still has them running. New nuclear is necessary but slow; small modular reactors (SMRs) have no commercial deployment anywhere in the world and realistic timelines put first European operations in the mid-2030s.
In the interim, gas fills the gap — whether stated openly or not. The dependency has not been eliminated. It has been rerouted and deferred.
V. The Fiscal Floor Drops Out
European governments are simultaneously subsidising EV adoption and depending on the fuel excise revenue that EV adoption destroys. This contradiction has received almost no serious policy attention.
Fuel excise duties function as a proxy road-use tax: you pay per kilometre driven, roughly, because fuel consumption correlates with distance. As EVs displace combustion vehicles, the proxy breaks down. Road infrastructure costs remain; the revenue base erodes. Across the EU, a cluster of medium-income economies — Poland, Hungary, Greece, and others — collect between 10-12% of total tax revenue from fuel and specific consumption taxes, making them structurally more exposed than the Netherlands or Denmark, whose broader tax bases absorb the loss more easily. Turkey, at the extreme, collects over 20% of total tax revenue from such taxes — a fiscal vulnerability to EV adoption that has attracted almost no discussion.
The replacement toolkit exists. Road pricing based on kilometres driven prices road use directly, and is being piloted in Oregon and designed in detail in the Netherlands. Per-kilowatt-hour levies at public charging points are administratively simple, though home charging creates an auditing problem. Ownership and registration taxes based on vehicle weight are the bluntest instrument but require no new infrastructure. No European country is implementing any of these at meaningful scale, because taxing EVs while simultaneously trying to incentivise them is a political contradiction that no government has been willing to own. The window to design the replacement system is open. It will not remain so indefinitely.
VI. Clean on Paper
The foundational justification for the EV transition is environmental. Trading oil for electrons reduces carbon emissions, cleans urban air, and breaks the climate feedback loop. The argument is broadly correct over a long enough time horizon. Over the short and medium horizon the picture is considerably more complicated.
Manufacturing a battery pack embeds a significant carbon cost — the mining, processing, and electrochemical production involved is energy-intensive, and that energy currently comes predominantly from fossil sources. A new EV starts its life with a carbon debt relative to an equivalent combustion vehicle. Depending on the carbon intensity of the grid it charges from, breakeven on lifetime emissions takes anywhere from two to four years of driving. On a coal-heavy grid the calculation looks worse still. On a nuclear-dominated grid it looks much better. The EV is not inherently clean. It is as clean as the electricity that feeds it, and the electricity that feeds it is, for the foreseeable future, substantially fossil.
The environmental cost of extraction rarely appears in transition advocacy. Lithium mining in the Atacama Desert — one of the driest environments on earth — is causing documented hydrological damage to salt flat ecosystems and indigenous water sources. Cobalt mining in the DRC generates a human rights record that no European company would be permitted to replicate domestically. These costs are real and geographically concentrated in places that lack the political voice to put them on the agenda of the countries driving demand.
The most environmentally rational strategy — maintain existing combustion vehicles longer, electrify incrementally as grids clean up, avoid manufacturing new cars where the embedded carbon cost is not yet justified by the operational saving — is industrially and politically inconvenient. It generates no new car sales, requires no subsidies to administer, and produces no ribbon-cutting moments for politicians. The transition as currently structured serves European industrial policy as much as it serves climate policy, and conflating the two has made both harder to think about clearly.
None of this is an argument against electrification. The direction is right (though not everyone would agree6). The problem is the romanticism — the insistence on presenting a complex, costly, and geopolitically fraught industrial transition as a clean break, a morning of clarity after the fossil fuel night. The bill is real. It is denominated in cobalt from the Congo, in gas burned to power data centres, in neighbourhood transformers that will take a decade to replace, in fiscal systems that have not yet been redesigned for the world they are supposed to be building.
Paying it honestly is the precondition for paying it well.
This essay was prompted by a recent interview on the economics of electric mobility, which raised the industrial questions sharply and stopped precisely where the structural ones begin.
Disclaimer:
Written in collaboration with Claude Sonnet 4.6
The term “battery factory” covers two distinct manufacturing steps that are often conflated. Cell manufacturing — the electrochemical process producing the individual cylindrical, pouch, or prismatic units containing anode, cathode, and electrolyte — is the technically demanding, capital-intensive step where the chemistry expertise resides. Battery pack assembly, by contrast, takes finished cells and combines them into the larger unit that goes into the vehicle, adding thermal management, electronics, and housing. European manufacturers have long performed pack assembly domestically while sourcing cells from Asia — which can appear in reporting as local battery production while the strategic dependency remains entirely at the cell level. Northvolt was specifically attempting to master cell manufacturing, not merely pack assembly, which is why its failure was more consequential than a generic factory closure.
The battery supply chain runs from raw material extraction through several distinct processing stages before reaching the cell factory. Mining extracts lithium, cobalt, nickel, manganese, and graphite from ore or brine. Refining and processing converts raw ore into battery-grade chemical compounds — lithium hydroxide, cobalt sulphate, nickel sulphate, purified graphite — the step China dominates most completely and which is hardest to replicate quickly. Active material production synthesises these refined chemicals into cathode and anode powders with precise electrochemical properties. Component manufacturing produces the separator films, electrolyte solutions, and copper and aluminium foil that go into each cell alongside the active materials. Only then comes cell manufacturing. Northvolt’s specific failure was not just at the cell manufacturing step but in its inability to secure competitively priced upstream material supply, without which the cell factory’s economics could not close.
Graphite is the least discussed of the battery material dependencies, yet arguably the most complete Chinese monopoly in the entire supply chain. Every lithium-ion battery cell requires a graphite anode — the negative electrode into which lithium ions insert themselves during charging. For this purpose, ordinary graphite must be processed into a spherical form: natural graphite flakes are milled into rounded particles of consistent size, then purified to carbon content above 99.95%, then coated with a thin carbon layer to improve electrochemical performance. This spherical graphite processing is the bottleneck, not graphite mining itself — deposits exist across Africa, Canada, and elsewhere. China controls roughly 95% of spherical graphite processing capacity globally. In late 2023, China imposed export controls on graphite, requiring licences for shipments abroad — a move that received a fraction of the coverage given to semiconductor export controls despite affecting every electric vehicle battery manufactured outside China. No Western country has meaningful spherical graphite processing capacity at commercial scale.
Transformers — the devices that step voltage up or down at every transition point in the electrical grid — are large, precision-engineered, and manufactured by a small number of specialist firms globally. A large power transformer can weigh several hundred tonnes and requires 12-18 months to manufacture under normal conditions. With every major economy simultaneously expanding grid infrastructure to accommodate renewables, EV charging, and data centre demand, order books are full across all major manufacturers and delivery times have stretched to 18-36 months or beyond for the largest units. Unlike semiconductor fabs or battery plants, transformer manufacturing capacity cannot be rapidly scaled — the bottleneck is as much skilled human expertise as physical plant.
The irony operates on two levels. First, the equipment being used to build Europe’s new energy infrastructure — transformers, inverters, grid-scale battery storage systems — is predominantly manufactured in China, meaning the construction of energy independence is itself a Chinese supply chain event. Second, and more durably, this dependency is harder to unwind than a fuel contract. A gas supply agreement can be renegotiated or redirected within months to years; a transformer installed in a substation will sit there for 30-40 years. If the geopolitical relationship with China deteriorates after the infrastructure is in the ground, the hardware cannot be quickly replaced. Europe is simultaneously attempting to escape one set of energy dependencies and installing, at the foundation level of its replacement energy system, a new set — not in contracts but in concrete and steel. The escape and the entrapment are being built at the same time, with the same budget, through the same procurement decisions.
The lifecycle carbon case for EVs rests on three variables that are less certain than the standard advocacy suggests. First, grid decarbonisation trajectory: the carbon saving per kilometre driven depends entirely on how clean the electricity is, and Section IV has shown that gas will fill a significant share of European generation capacity for at least 15-20 years. Second, the electricity availability assumption: the lifecycle argument assumes sufficient clean electricity exists to charge the fleet. With AI data centre demand and EV charging competing for the same grid capacity, and grid expansion lagging both demand curves simultaneously, this is a projection built on policy targets rather than guaranteed infrastructure. Third, manufacturing emissions vary enormously by factory location and energy mix — a battery cell produced in a coal-powered Chinese factory carries a very different embedded carbon cost than one produced on hydropower. On the most favourable assumptions — a rapidly decarbonising grid, sufficient generation capacity, and clean manufacturing — the lifetime carbon arithmetic favours the EV clearly. On less favourable assumptions, which current evidence suggests are more realistic in the medium term, the case is real but considerably more qualified than public discourse acknowledges.