The Powertrain Race Was Never a Technology Contest: Why the Battery Already Won as a Business Model

There is a debate raging across Germany and much of Europe about how cars should be powered in the future. Battery-electric. Hydrogen. eFuels. HVO100. The framing is almost always the same: which technology is best? Which one is cleanest, most convenient, most German? The word you hear most often is Technologieoffenheit – technology-openness, the noble-sounding demand that we keep all options on the table until the engineering is settled. What almost nobody argues about is the powertrain business model.

The framing itself is the mistake. The question is not which powertrain technology is best. The question is which business model wins – and where it stands on the two curves that decide such contests: the adoption curve on the demand side, which determines whether enough customers have actually switched, and the learning curve on the supply side, which determines whether the technology gets cheaper fast enough to matter. And once you ask it that way, the answer is no longer open. It has, in the sense that matters for anyone making a capital-allocation or career decision, already been decided.

The same argument settled an earlier contest. In the digital economy of the late 1990s the consensus held that the established media giants would converge and win. They did not. New entrants with new business models did – which is what the 2001 dissertation that first treated the business model as a distinct unit of analysis had predicted. The case here is a very physical industry, but the tools are the same, and they long predate this piece.

The Wrong Unit of Analysis: Why the Powertrain Debate Compares Products Instead of Business Models

The mistake in the public debate is a category error. People treat the powertrain question as a comparison of products – this engine versus that engine, this fuel versus that fuel – the way you would compare two washing machines on a shelf. But a car is not a product sold on a shelf. It is the visible tip of a business model: a value proposition delivered through a value-creation architecture, financed by a revenue model, carried by people and values. When one such system displaces another, the winner is rarely the better product. It is the better-configured system.

Beyond the Single Business Model: Why an Ecosystem Is Really an Architecture of Interlocking Business Models

There is a second level to the category error, and it matters more than the first. When people do move past the product to talk about the business model, they usually still picture a single company’s business model – BYD’s, or Volkswagen’s. But what is actually being decided here is larger than any one firm. It is a whole architecture of interlocking business models, and the winner is the architecture that coheres, not the company that optimizes.

Think about everything that has to work together for the battery-electric answer to deliver affordable mobility. Raw-material extraction and refining. The machine-building industry that produces the equipment to make cells. The chemistry to process and refine battery materials in the first place. New tiers of suppliers reorganizing around the automotive industry. Charging infrastructure. And new regulation for how battery safety is assured and how cells are recycled at end of life. Each of these is its own business model, with its own customers, its own value-creation architecture, its own revenue logic. None of them succeeds alone. This is what is usually called an ecosystem today, and the word is not wrong so much as too comfortable: it describes that many players somehow belong together, while leaving out the only thing that decides the outcome – how they depend on one another. An architecture of interlocking business models is the more demanding term, because it forces you to name the interdependencies rather than gesture at them. The tipping point is not one business model beating another – it is one system of mutually reinforcing business models reaching the coherence and scale at which it out-delivers the incumbent system. This is why the definition of a business model used here insists on interdependence rather than a list of parts: here the interdependence runs not only between the elements of one model, but between whole models stacked into an architecture. Synthetic fuels and hydrogen-for-cars lose not merely as products, and not merely as single business models, but as attempts to swap one component inside the old architecture while the new architecture is being assembled around them, piece by piece, into something that hangs together.

Two Theories That Explain Why Markets Tip: Rogers and Abernathy-Utterback

To see why the race is already decided, you need two ideas that have been sitting in the innovation literature for decades. Both are worth more than most of what passes for strategy commentary.

The first is Everett Rogers‘ diffusion of innovations (Rogers, 2003, first published 1962). Rogers’ enduring insight is that adoption is not a smooth ramp driven by how good a product is. It follows an S-curve driven by social structure. A small group of innovators and early adopters goes first. Then, if the innovation crosses a threshold of social proof, the early and late majority follow in a rush. That threshold is not a metaphor. The 2001 dissertation derived it formally from the economics of word-of-mouth: one additional recommendation per customer – the difference between five and six – can produce a roughly forty-fold difference in the resulting customer base. Below the threshold, stagnation. Above it, self-sustaining growth. The tipping point is a mathematical fact, not a mood.

The second is William Abernathy and James Utterback‘s theory of dominant design (Abernathy & Utterback, 1978). Their observation was that a young industry begins in a fluid phase: many competing product designs, rapid experimentation, no settled answer. Then a dominant design emerges – and the moment it does, competition flips. It shifts from wild product variety to process refinement, scale, and cost. The number of players collapses in a shakeout. The industry enters its specific phase. I already used this lens on my blog to explain how DeepSeek unsettled the AI industry, which is still in its fluid phase. The automotive powertrain is in the opposite situation: it is leaving its fluid phase right now.

One move extends the theory, and it is the same one made in 2001: Abernathy and Utterback located the dominant design at the level of the product. Lift it to the level of the business model. What consolidates is not merely a preferred configuration of components. It is a preferred configuration of value creation – the whole system. That dissertation predicted that in digital markets, “the selection process among competing business models will produce dominant designs,” which would create temporary oligopolies built on demand-side network effects and supply-side scale effects. That was written about eBay and Amazon. It reads, today, as a description of what is happening to the car.

Put the two theories together and you get the anatomy of a tipping point. A market tips when demand and supply take off at the same time – when Rogers’ adoption threshold is crossed on the demand side and Abernathy-Utterback’s dominant design locks in on the supply side. Note where on the curve this happens: not at the inflection, where growth is already fastest, and certainly not where growth flattens, but earlier, at the point of maximum acceleration – the moment slow change turns into self-sustaining growth. That is the moment worth recognising, because it is the last one at which the outcome still looks open. Each side is driven by effects that reinforce themselves. That is the engine of the whole story, and I will now show you exactly where those self-reinforcing effects sit, using the four elements of the business model as a diagnostic grid.

Where the Self-Reinforcing Effects Sit: The Four Elements of the Winning Powertrain Business Model

The value of a framework is that it tells you where to look. Run the powertrain contest through the four elements and the reinforcing loops appear in specific, nameable places rather than as a vague sense that “electric is winning.”

The Value Proposition: Why Demand Has Already Crossed the Threshold

Two distinct forces are tipping demand, and it is worth keeping them apart because they operate differently.

The first is Rogers’ adoption threshold. We are no longer in the innovator-and-early-adopter phase of electric mobility. In 2025, global electric-car sales passed 20 million and reached roughly 25% of all new cars sold, according to the IEA’s Global EV Outlook 2026. In China, new-energy vehicles reached about 48% of the market. In Norway, battery-electric cars were around 96% of new sales. These are not the numbers of a visionary fringe. They are the numbers of the early and late majority arriving in force – the steep part of the S-curve. Crucially, Rogers tells us why this accelerates: people do not adopt because a spec sheet is superior. They adopt because trusted peers already have, and report that it works. Every neighbour’s electric car that does not strand its owner is a recommendation. This is what I call the building of consumption knowledge: as more people learn how charging actually fits into life, the perceived cost of adoption falls for everyone still watching. The barrier was never only money. It was the time and uncertainty of learning a new routine – and that barrier erodes with every adopter.

The second force is a genuine network effect on the infrastructure side. Charging is the obvious case: the denser the fast-charging network, the lower the friction for the next buyer, which raises demand, which justifies more chargers. Public charging connectors grew about 28% in 2025 to some 6.7 million worldwide (BloombergNEF, 2026). But the reinforcing loop runs through the used-car market and residual values too: more new EVs today means a deeper, more trusted second-hand market in three years, which lowers the entry price for later, more price-sensitive adopters. Each of these loops feeds the value proposition of affordable, frictionless mobility – and each gets stronger precisely because it is already growing.

Precision matters here. This infrastructure effect is a real network effect, but it is not a platform network effect of the kind that made the tech giants near-invincible. A charging network is closer to physical logistics than to a two-sided market. Confusing the two would be exactly the category error this piece warns against: local logistics is not a platform. The demand-side tipping of EVs is real, but its dominant driver is Rogers-style adoption plus infrastructure, not Metcalfe’s law. The truly decisive self-reinforcing effect lives on the other side of the business model.

Why Hydrogen and eFuels Never Crossed the Adoption Threshold

Before I turn to the supply side, apply the exact same Rogers lens to the three alternatives. It is revealing, because each fails – or survives – at a different point in the business model, and the differences matter.

Hydrogen is watching its network effect run in reverse. The infrastructure loop that helps batteries is, for hydrogen passenger cars, actively unwinding. Germany’s public hydrogen refuelling network fell from more than 90 stations to around 50, with roughly 40% of the remaining car-focused sites closing by the end of 2025 (Clean Energy Wire, 2026; Hydrogen Insight, 2025). Austria dismantled its entire public network (CleanTechnica, 2025). The numbers underneath are brutal: a German hydrogen station serves on the order of eight cars a day, against 200 to 300 for a petrol station (CleanTechnica, 2026). This is Rogers in reverse: too few vehicles make stations unprofitable, closures make the proposition worse for the next buyer, which suppresses vehicles further. A self-reinforcing loop pointed downhill.

Put the demand next to the battery and the mismatch is not close. Worldwide, just 16,011 hydrogen passenger cars were sold in 2025 – and even that modest figure was flattered by year-end tax effects in China; the global fuel-cell market peaked at 20,704 units back in 2022 and has fallen every year since (electrive, 2026; SNE Research via Driving Hydrogen, 2025). Europe managed 485 hydrogen cars in the first half of 2025. In Germany, registrations fell by nearly 70% in 2023. Against more than 20 million electric cars sold globally in the same period, hydrogen for passenger cars is not an early-stage competitor climbing the S-curve. It is a technology that reached its tipping point, failed to cross it, and is now sliding back down.

The scope of that claim is often mangled in both directions. Nothing here is an argument against hydrogen as such. Hydrogen is indispensable, and will remain so, in the places where nothing else does the job: ammonia and fertiliser production, refinery hydrocracking and desulphurisation, methanol, and the direct reduction of iron for green steel. Michael Liebreich’s Clean Hydrogen Ladder is the most useful way to hold both thoughts at once: it ranks hydrogen use cases from A, where there is no alternative, down to G, which Liebreich calls the Row of Doom. The ranking weighs cost, efficiency, safety, critical-mineral availability and, decisively, whether a cheaper or better alternative already exists. Fertiliser and hydrocracking sit at the top. Cars sit at the bottom. That is not a verdict on the molecule. It is a verdict on one application of it – the one where a directly electric alternative is cheaper, more efficient and already deployed at scale. Hydrogen loses the car. It wins the fertiliser plant.

But a second warning belongs next to the first, and it is one the hydrogen-for-industry camp likes even less. Being indispensable is not the same as being cheap, and it is not the same as being on a self-sustaining curve. Clean hydrogen has barely started its own descent. Low-emissions production reached roughly one million tonnes in 2025 – the first time the category crossed one percent of total hydrogen output – and installed electrolysis capacity doubled to just over four gigawatts (IEA, Global Hydrogen Review 2026). Those are the numbers of a technology at the very top of its learning curve, not one walking down it. And the direction of travel in 2025 was backwards: new final investment decisions fell below 0.8 Mt a year after two years at around 1 Mt, the project pipeline to 2030 shrank by 10 Mt to 27 Mt, and more than 100 GW of announced electrolysis capacity will lose any chance of operating before 2030 unless it is committed by the end of 2027. The IEA’s own wording is blunt: electrolyser manufacturing is entering a consolidation phase because the market is developing too slowly. The cost gap explains why. Unsubsidised green hydrogen in Europe runs at roughly €4 to €7 per kilogram against industrial buyers who will pay under €3, and in the European Hydrogen Bank’s second auction seven projects representing 1.88 of 2.33 GW simply withdrew.

This is the same diagnosis I applied to the car, pointed at the industrial case: the volume that walks a technology down its cost curve is not yet there. The difference is that in fertiliser, refining and steel there is no cheaper alternative waiting – so the learning curve has to be climbed rather than abandoned. That is an argument for concentrating scarce hydrogen effort exactly where it is unavoidable, and against spreading it thinly across applications that a wire and a battery already serve better. Every kilogram burned in a passenger car is a kilogram that does not descend the curve where it actually matters.

Here the hydrogen advocates have a ready answer: never mind the private market, the state will manufacture the demand. Germany is doing exactly that. BMW’s iX5 Hydrogen programme – a pilot fleet of under 100 vehicles heading into deliberately low-volume series production in 2028 – is backed by €273 million in public funding, €191 million from the federal transport ministry and €82 million from Bavaria. This is worth taking seriously, because it is precisely the kind of move that sounds like it should work and does not. You cannot decree a Rogers tipping point. The adoption threshold is crossed by social proof spreading through peer groups – by real customers persuading other real customers – not by a few hundred or few thousand state-commissioned vehicles. Subsidised volume of that size does not trigger the self-reinforcing loop; it simulates critical mass without producing it, which is why the counterfeit demand evaporates the moment the subsidy stops. Tellingly, BMW’s own spokesperson concedes the real condition: the hydrogen pump price must reach parity with diesel for the vehicle to succeed commercially. That is a learning-curve problem – the very problem the programme’s volumes are far too small to solve. And the language around it – “only innovation, not bans”, “technology-openness” – is the same preservation reflex the HVO100 post flagged, now wearing a hydrogen badge. A subsidy can buy a pilot fleet. It cannot buy the volume that walks a technology down its cost curve, and it cannot manufacture social proof.

There is a further, deeper problem in the value-creation architecture itself, and it cuts to the core of what makes a business model defensible. The scale effect fails at the most important component: the fuel cell. BMW does not own it. The first-generation stack was supplied entirely by Toyota; in today’s iX5 Hydrogen pilot fleet, BMW engineers the system but sources the individual cells from Toyota; and the third-generation stack for 2028 is being co-developed with Toyota precisely to share the immense R&D cost. These vehicles are, in BMW’s own description, hand-built and expensive – the fluid phase of Abernathy and Utterback, a workshop and not a learning curve. In the language of my book, this is the absence of what I call UBM, the uniqueness of a business model: durable advantage lives in the interdependence of several components you actually own. A borrowed core component is, as I put it in the online-marketing post, one borrowed block on someone else’s land. The battery architecture scales its core component across tens of millions of units and rides the cost curve down. The fuel-cell architecture builds its core component by hand, splits it across two companies, and produces it in the low hundreds. Even the co-development is an admission that neither partner alone can reach the volume that scale economics require.

HVO100 is the instructive opposite: the demand is there. Customers would happily pump renewable diesel. It is a drop-in fuel, it works in existing engines, it needs no new consumption knowledge and no new infrastructure. On the demand side, HVO100 passes every Rogers test that hydrogen fails. And yet it will not carry the mass market – not because customers reject it, but because, as the HVO100 post argued, the constraint sits entirely in the value-creation architecture: there is not, and will not be, enough sustainable feedstock to produce it at the scale of the car fleet. It is a trickle, not a stock. HVO100 proves a point that matters for every business model: a value proposition the customer loves still fails if the architecture cannot deliver it at scale. Demand is necessary. It is not sufficient.

The Value-Creation Architecture: The Learning Curve Is the Real Engine

Here is the heart of the matter, and it is where the most rigorous science in this whole debate lives.

In 1936, the aeronautical engineer Theodore Wright observed that the cost of producing airplanes fell by a consistent percentage every time cumulative production doubled. This is Wright’s Law, the learning curve. It has since been shown to hold across dozens of technologies. For lithium-ion batteries the evidence is now overwhelming: Our World in Data documents that since 1998, every doubling of cumulative battery production has cut the price by roughly 19% – a rate close to that of solar panels, sustained across a 27-million-fold increase in cumulative output. BloombergNEF puts the learning rate near 18% and recorded average pack prices around or below $100/kWh – the level at which electric and combustion cars converge on cost.

This is the supply-side self-reinforcing loop, and it is more powerful than any marketing effect: every battery produced makes the next one cheaper. Falling cost expands demand; expanding demand drives cumulative volume; cumulative volume drives the next cost reduction. Scale is not a one-time advantage here. It is a flywheel. And it does not stop at manufacturing. The same accumulation happens in the know-how around the battery – in production engineering, in charging infrastructure, in servicing and repair – and, critically, in research. Sustained volume funds sustained R&D into new cell chemistries such as lithium iron phosphate and sodium-ion, which do something strategically decisive: they move or dissolve the bottleneck. A chemistry that needs less lithium or cobalt loosens the very raw-material constraint that skeptics point to as the ceiling on batteries. The learning curve, in other words, is not just lowering cost. It is dismantling its own limits.

How Cordless Drills and Mobile Phones Built the Car Industry’s Nemesis

Now ask the question that the powertrain debate almost never asks: why was the battery already so far down its curve when the car race began? The answer is that it did not start with the car. Lithium-ion cells have been in commercial mass production since 1991, and for their first two decades the volume came from somewhere else entirely: laptops, camcorders, cordless drills, toys, and then, overwhelmingly, mobile phones. By the time the first serious electric cars arrived, the industry had already accumulated many doublings of cumulative production in a market that had nothing to do with mobility. MIT research by Ziegler and Trancik documents the result: lithium-ion prices fell by roughly 97% over three decades of commercialization, long before automotive demand mattered.

There is an irony here that ought to be uncomfortable in Stuttgart, Munich and Wolfsburg. While Europe’s premium carmakers spent those same decades perfecting combustion engineering – the throttle response, the exhaust note, the chassis tuning that justified the badge – their eventual competitor was quietly compounding its cost advantage inside power tools and handsets. Nobody in the automotive industry watched that market, because nobody in the automotive industry considered it a market at all. It was consumer electronics. It was toys. It was, by the standards of a company that builds engines, unserious. That is exactly why it was dangerous: the threat matured out of sight, in an arena the incumbents had no reason to monitor.

And there is a second, sharper reversal buried in this. For most of automotive history, the carmakers were the feared customer. Volkswagen at ten million vehicles a year was a buying power that suppliers organized their entire existence around, and everyone in the industry knew how that power was used. But that dominance rested on a quiet assumption: that the car industry was the volume in its component markets. For a crankshaft or an exhaust system, it was. For a battery cell, a chip or a sensor, it is not.

Look at TSMC, the Taiwanese foundry that manufactures the chips almost every modern car depends on. Its largest customer is Nvidia, at roughly 22% of revenue; Apple is second at around 18 to 25%; the top five customers together account for an estimated 65 to 70% of the total (Nvidia overtaking Apple; TSMC customer concentration). And the entire global automotive sector – every carmaker on earth, combined – sits at about 5% of TSMC’s revenue, behind high-performance computing at roughly half and smartphones at around a third (TSMC segment breakdown). Read that again: the whole of the world’s car industry is a smaller customer than Nvidia alone. Smaller than Apple alone.

Automotive accounts for only 8 to 11% of global semiconductor demand overall (Oliver Wyman); as one analyst put it plainly, at around a tenth of the market, carmakers are “by no means a high priority” from the semiconductor industry’s point of view (EE Times). The learning curve in those components is set by phones, laptops and now AI data centres. When memory makers must choose, they allocate capacity to high-bandwidth memory for servers, where the same wafer earns far more, and let automotive pay up or wait. Ten million cars was never the point. Ten million cars only conferred power while the car industry was the biggest buyer in the room – and in the components that now decide the car, it no longer is. The industry that once dictated terms to its suppliers now queues behind the very consumer electronics it never took seriously.

The crucial point is what transfers. The chemistry diverges: a phone cell and a car cell often use different cathodes. But the manufacturing process is nearly identical – the same slurry mixing, the same electrode coating, the same calendering rolls, the same winding or stacking, the same electrolyte filling in dry rooms, the same formation cycles. Improvements in coating speed, yield, drying energy, and automation are chemistry-agnostic: they help the car cell exactly as much as they helped the phone cell. That is why China’s two battery giants are not newcomers to the technology. BYD was founded in 1995 as a maker of rechargeable cells for handsets and only bought a carmaker in 2003. CATL is a 2011 spin-off of ATL, which had spent a decade making cells for laptops, MP3 players, and then smartphones for Samsung and Apple. Neither is a car company that learned to make batteries. Both are battery companies that later decided to put wheels underneath their cells – and they arrived at the automotive starting line with more than a decade of accumulated process learning already behind them.

Hydrogen and eFuels never had such a market. There was no consumer industry that spent thirty years mass-producing electrolyser stacks or synthesis reactors and grinding down their costs in the process. No one ever put a PEM stack in a laptop or a Fischer-Tropsch reactor in a cordless drill. The challengers must therefore earn every single doubling from scratch, inside the one application they are trying to win – while the battery arrives with three decades of borrowed volume already compounding in its favour. This is not a detail. It is the reason the two curves sit where they sit.

The Revenue Model: The Cost Gap, and the Honest Objection

Every element so far points one way, which makes it the moment to state the strongest counter-argument and answer it. Getting this wrong is the most common mistake in the debate, and an easy one to make before looking closely.

The lazy version of my argument would be: “batteries have a learning curve and hydrogen and eFuels do not.” That is false. Hydrogen electrolysers have a well-documented learning curve. A European dataset on green-hydrogen production finds PEM electrolyser investment costs falling by roughly 15% to 17% per doubling of cumulative capacity; IRENA derives learning rates in the 16-21% range by analogy to solar. eFuels, too, would come down with scale. The problem is not that the challengers cannot learn. It is subtler, and it is pure economics.

A learning curve only advances if you win the volume that feeds it. Hydrogen and eFuels sit today on a far higher cost position, with far fewer cumulative doublings behind them. To catch up, they would need to capture enough demand to keep doubling their cumulative output – but the battery is already absorbing that demand, running down its own curve, and pulling further ahead. The challengers are not losing because they cannot learn. They are losing because the rival is eating the volume that learning requires. This is precisely the dynamic that the economist W. Brian Arthur formalized as increasing returns to adoption (Arthur, 1994): in the presence of self-reinforcing feedback, a technology that gains an early lead is chosen by ever more adopters, until the market locks in – not necessarily on the theoretically best option, but on the one that got the momentum. Path dependence, lock-in, the QWERTY keyboard that never left. The lead compounds.

And there is a floor beneath the challengers that no amount of learning can lift, because it is not a manufacturing cost – it is thermodynamics. Hydrogen and eFuels are trapped by their conversion chain. To make an eFuel you take electricity, split water into hydrogen, combine it with captured CO₂, synthesize a liquid, then finally burn that liquid in an engine that is only about 35% efficient. Every step sheds energy. The ICCT estimates that roughly half the input energy is lost just in converting electricity to liquid fuel; industry analyses note that a litre of eFuel needs around 25 kWh of electricity to produce but contains only about 10 kWh. The battery path skips almost all of this. You produce the battery once – that is where the learning curve does its work – and thereafter it stores electricity and delivers it through an electric motor that is roughly 80% efficient, straight into the benefit the customer actually wants: motion.

This is the same stock-versus-flow distinction the HVO100 post drew, now on the level of energy. The conversion chain of hydrogen and eFuels is a flow that leaks value at every stage. The battery holds the state and releases it with very little loss. One architecture spends most of its energy fighting its own process. The other delivers the customer benefit directly. No learning rate repeals the second law of thermodynamics.

There is one more test, and it is the cleanest of all, because it removes the battery from the picture entirely. Consider aviation. Here there is no battery competitor absorbing the volume – for long-haul flight, eFuels (e-kerosene) are widely regarded as the only realistic path to deep decarbonization, and unlike biofuels they face no feedstock ceiling. If “the battery is stealing the volume” were the whole explanation, e-kerosene should be racing down its learning curve in the one place it has the runway to itself. It is not. The European Union Aviation Safety Agency puts 2024 power-to-liquid eFuel production costs around €7,695 per tonne – roughly ten times the cost of fossil kerosene – and both the ICCT and Carbon Direct conclude that even as costs fall, eFuels are expected to stay far above fossil jet fuel through 2050. The World Economic Forum calls it plainly a chicken-and-egg problem: airlines signal demand, yet supply stays tiny and prices stay two to three times higher. The lesson is decisive. Even where synthetic fuel is the only option and has no rival eating its volume, the sector cannot generate enough demand fast enough to walk the cost position down to parity. The conversion-chain penalty is simply too deep a hole to climb out of on a learning curve alone. This is not an argument against eFuels for aviation, where they may be indispensable. It is the final piece of evidence that the constraint is structural, not merely competitive.

The Company Spirit: Why the Incumbents Cannot Simply Follow

If the dominant business model is visible, why do the incumbents not just adopt it? This is the fourth element – the human one, team and values – and it is the one I care about most, because transformation is never a technical problem. It is a human one.

Start with the dynamic that traps them, because it is more specific than “resistance to change.” Clayton Christensen‘s Innovator’s Dilemma describes a precise trajectory. A disruptive technology usually arrives worse on the dimensions the best existing customers care about. Early electric cars were exactly that: short range, slow to charge. So the most demanding customers rightly rejected them, and the incumbents, listening to those customers, rationally stayed with the combustion engine. Christensen’s twist is what happens next: the disruptive technology improves along its own steep curve until it becomes good enough – and then better than good enough – on the very dimensions that once held those customers back. At that point even the most demanding buyers switch, and they switch fast.

That is not a hypothetical here; it is happening on the one axis that was the last refuge of the combustion argument: charging speed. Older EV batteries charged at roughly 2C – the rate at an average 150 kW session. The latest cells reach 10C to 12C in the early phase of charging, a five- to six-fold improvement, according to the IEA’s analysis of ultra-fast-charging batteries. In practice that means BYD’s second-generation Blade battery charging from 10% to 70% in five minutes, and CATL’s third-generation Shenxing going from 10% to 98% in about six and a half minutes – close to the time it takes to fill a tank. This is the crucial point that pure cost curves miss: the battery is not only getting cheaper (Wright’s Law), it is getting better on the dimension that demanding customers used to point to as the reason not to switch. Christensen did not dwell on the mechanism of that improvement. We now can name it: the same learning curve, funded by the same volume. Range and charging time were the incumbents’ last defensible ground, and both are being taken.

Now the human question: if the trajectory is this clear, why do incumbents still hesitate? Here the familiar idea of the efficiency trap adds to Christensen rather than repeating him. The trap is not that firms are lazy or blind to the disruptor. It is that efficiency measures how well you do something and is silent on whether the something is still worth doing – the old distinction between doing things right and doing the right thing, which goes back at least to Drucker. Applied to the business model, as in the earlier post on why great online marketing is not a defensible business model, it cuts deep: a firm is often at its most productive precisely when it is closest to obsolescence, because decades of optimization have made the old model magnificent. The most perfectly optimized diesel plant in Europe is not an asset in an electric world; it is a sunk cost that makes the necessary transition harder to justify, because abandoning it means writing off the very thing the organization is proudest of. Christensen, Kaufman and Shih sharpen the financial side of this in Innovation Killers: the incumbent weighs a new plant against the marginal cost of running its already-depreciated assets, while the attacker faces only the full cost of the new – so the numbers rationally, and fatally, favour leveraging the old. The better you have optimized the old architecture, the more it argues against leaving it. That is the trap: not incompetence, but excellence pointed at the wrong target. The SANY electric-truck case makes it concrete: where the cost gap to diesel collapsed from 3× to about 20% in four years – driven not by an engineering breakthrough but by the cost structure that scale enables. The perfect carriage still lost to the first car.

Complicated Is Not Complex: Why the Old Tools Fail Here

There is a deeper reason the incumbents reach for HVO100 and eFuels, and it is a failure of diagnosis before it is a failure of will. It is worth naming with Dave Snowden’s Cynefin framework, combined with the Three Horizons.

Optimizing a diesel powertrain is a complicated problem. Cause and effect are known; expertise and planning deliver reliable gains; this is the work of Horizon 1, the existing business. eFuels and hydrogen-for-cars are, in large part, a Horizon 1 move: a rational attempt to preserve the installed base by swapping the molecule while keeping the entire value-creation architecture – engines, refineries, filling stations, dealer networks – intact. Within the logic of the complicated, it looks sensible. Keep the assets. Change one input.

But the powertrain transition is not a complicated problem. It is a complex one – a Horizon 2/3 shift where the winning configuration cannot be known in advance by analysis, only discovered by moving. And the tools built for the complicated – efficiency programs, cost optimization, defending the installed base – are not merely weak in the complex domain. They are actively dangerous, because they pour resources into perfecting a design that the market is in the process of abandoning. This is the trap the mechanical-watch industry escaped only by redefining its value proposition entirely: when quartz made cheap, accurate timekeeping universal, mechanical watches survived not by competing on accuracy but by selling a different benefit – status, craft, permanence. HVO100 and eFuels, by contrast, try to sell the old benefit through a more expensive architecture. That is optimization aimed in exactly the wrong direction.

The Learning Curve Can Run Backwards: The Incumbent Is Not Safe Either

So far the story has been about challengers who cannot get down their learning curve fast enough. But there is a mirror image that the incumbents should find far more disturbing, and it follows from the same economics. If a learning curve runs down with rising cumulative volume, it can also run up with falling volume. Scale economies work in reverse. Call it the de-scaling effect.

The entire fossil value chain is a scale machine: exploration and drilling equipment, refineries, pipeline and tanker logistics, the filling-station network, and the engine-manufacturing base that feeds it. Every one of these was built for, and is only economic at, enormous throughput. A refinery is a high-fixed-cost asset that needs to run near capacity; a filling-station network is only worth maintaining if enough cars still pull in; an engine plant only earns its tooling back over millions of units. Now run the volume down, as electric mobility is already doing on the demand side, and the cost per remaining unit does not hold steady. It climbs. Fixed costs spread over fewer barrels, fewer litres, fewer engines. This is a negative scale effect propagating across all four elements of every business model along the chain at once – value proposition, architecture, revenue model, and the people who staff it. We saw the miniature version already in this post: Germany’s hydrogen stations closing not because hydrogen got worse but because throughput fell below the level that makes a station viable. The same logic, applied to the vast fossil chain, is a slow-motion version of the same unwinding.

Today this is invisible, and it is worth being precise about why it is invisible, because the reason is also the vulnerability. The chain is held up by extraordinary margins at its cheapest source. Saudi Aramco’s lifting cost is about $3.50 per barrel; its CEO cites extraction costs nearer $2. Against a 2025 realized oil price of roughly $69 per barrel, that is a contribution margin above $65 a barrel – a gross margin north of 90% at the wellhead. That river of cash from the lowest-cost fields is what subsidizes the survival of every more expensive link in the chain: the higher-cost shale and offshore production at $30 to $70 per barrel, the refineries, the retail networks. But note the trap hidden inside those numbers. The production cost is $3.50; the fiscal break-even that Saudi Arabia’s national budget needs is estimated by the IMF at around $91 to $96 per barrel, far above today’s price. The cheapest producer on earth can keep pumping almost indefinitely on a per-barrel basis – but the rentier economy built on top of that oil assumes both high prices and high volumes. Remove the volume, and the incumbent that looks unbreakable is revealed to be leaning on the same scale logic as everyone else, only from the comfortable end of it. The perfectly optimized diesel plant, and the perfectly optimized oil economy behind it, share a vulnerability: both were built for a volume that the tipping point is quietly taking away.

Winning the Powertrain Business Model Is Not the Same as Owning It

There is an uncomfortable coda here, and it is the part that should worry European readers most. Recall the point about BMW not owning its fuel cell: the core component of the hydrogen car comes from Toyota. Intellectual honesty demands the same question of the winning architecture. Does Europe own the core component of the battery car? Largely, it does not.

China controls more than 75% of global lithium-ion cell production and an even larger share of the cathode and anode materials upstream of it; CATL alone holds close to 38% of the global market and over 40% in Europe (S&P Global, 2025). Europe’s attempts to build a home-grown champion have mostly faltered: Northvolt filed for bankruptcy in 2025, and the European gigafactory pipeline has been cut sharply by cancellations and delays (Oxford Institute for Energy Studies, 2025). Worse, Europe barely produces LFP cells at all – the cheap, safe chemistry that now dominates the mainstream market – so the LFP capacity appearing on European soil is largely CATL’s in Hungary and the CATL-Stellantis joint venture in Spain. Even BMW sources the cylindrical cells for its Neue Klasse from China’s EVE Energy.

Hold the two facts side by side and the strategic tragedy is complete. On the losing architecture – hydrogen for cars – Europe does not own the core component, but that hardly matters, because that architecture loses anyway. On the winning architecture – the battery – Europe also does not own the core component, and that matters enormously, because this is the model that will carry mobility for the next generation. This is the deeper failure beneath the powertrain debate. While the energy went into defending a dying architecture with synthetic fuels and hydrogen cars, the value-creation architecture of the winning model – the cell, the chemistry, the materials – was being built somewhere else. Winning the argument about which model dominates is not the same as owning the model. Europe is on course to be a customer of the dominant business model rather than an author of it. That is not a technology problem. It is a business-model problem, and it was decided by where the scale, and therefore the learning curve, was allowed to accumulate.

And remember the frame from the start of this piece: this is not one component but a whole architecture of interlocking business models – materials, machine-building for cell production, chemistry, supplier tiers, charging, recycling, regulation. Owning the winning model means participating across that architecture, not bolting on a single gigafactory late and hoping to catch a learning curve that has been running for a decade elsewhere. A lone cell plant, dependent on imported chemistry and imported production equipment, is not ownership of the architecture. It is a tenant in someone else’s building. The strategic task was never to pick the winning technology. It was to build a coherent system of business models around it, early enough for the interdependencies to compound at home rather than abroad.

What Is Genuinely Still Open: Trucking, Aviation and Regional Divergence

Intellectual honesty requires drawing the boundary of the claim precisely, because the tipping point is not uniform across all of mobility, and pretending otherwise would be its own kind of bad strategy. Confront the brutal facts, as Jim Collins would say – including the ones that complicate your thesis.

Several things remain genuinely undecided, though fewer than the “technology-openness” camp likes to claim. Take heavy long-haul trucking, often called the hardest passenger-adjacent segment to electrify. Even here the direction is clearer than the debate suggests – and in one market it has already tipped. In China, over 450,000 zero-emission heavy trucks were sold in 2025, a 25% share up from 3% in 2021; in the tractor-trailer segment – the heaviest category, long considered the hardest to electrify – the zero-emission share reached nearly 30% (ICCT, 2025), and in December 2025 electric heavy trucks briefly outsold diesel for the first time. Tellingly for the dominant-design argument, that market is already consolidating: the top five zero-emission truck makers (including SANY) hold about 61% share, a maturity close to the diesel market’s. Europe is far earlier on the curve – electric trucks were roughly 4.5% of sales in 2025 – but growing about 60% year-on-year as the EU’s first CO₂ standards bite (ICCT, 2026), and Chinese makers including SANY are entering with long-range tractors. On cost, the IEA expects battery-electric trucks to undercut diesel on total cost of ownership in China and the EU by 2030, with fuel-cell trucks staying more expensive in both markets as megawatt charging erodes hydrogen’s refuelling-time advantage; a long-haul analysis by Energy Innovation finds battery-electric economically stronger even where fuel cells were supposed to win. The SANY electric-truck post documents the same faster-than-expected compression. But honesty cuts both ways: some peer-reviewed work still finds diesel cost-optimal across many ranges and weights, partly because a long-haul diesel engine runs near 45% efficiency against a passenger petrol engine’s 35% – so the electric efficiency advantage, though still real, is narrower here than for cars. The segment is moving the same way, just later outside China and with genuine open questions about charging logistics and payload. Aviation and shipping are less settled still; their energy density and duty cycles are the cases where synthetic fuels may hold a durable and legitimate role – not because they are cheap there, but because, as the aviation numbers above show, they may be the only physical option, cost premium and all. Regional divergence is real: China has tipped hardest and fastest, while Europe is tipping through regulation as its first CO₂ standards take effect (BloombergNEF, 2026). Grid capacity and charging build-out are real constraints that will pace the transition in some regions. And raw-material prices can bite: lithium prices in early 2026 ran more than double their level a year earlier, even while remaining far below the 2022 peak (IEA, 2026). None of this reverses the direction of travel for passenger cars. All of it should temper any claim that the transition is finished, uniform, or frictionless. The tipping point is reached for the passenger car. It is not reached everywhere, for everything, all at once.

Seven Business Model Lessons Every Industry Can Learn

Strip away the cars and the fuels, and this case leaves behind a set of tools you can point at your own industry the next time someone tells you a contest is still about which technology is best. That is the real payoff, and it is why I bother with cases at all.

1. Choose the right unit of analysis – and it is bigger than you think. The powertrain debate goes in circles because it argues about engines and molecules. Move up one level to the business model and the fog begins to clear. But move up one more level, to the architecture of interlocking business models – materials, machines, chemistry, suppliers, charging, recycling, regulation – and you see what is really being decided. Products compete on a shelf. Business models compete as systems. And systems of business models win or lose as whole architectures, on whether their interdependencies cohere and compound. Pick the frame that matches the contest, or you will answer the wrong question well.

2. Find the learning curve, and ask who is feeding it. Almost every technology has a learning curve. The strategic question is never “does it improve with scale?” – nearly everything does. It is “who is winning the cumulative volume that drives the improvement?” A challenger with a steeper potential curve still loses if the incumbent alternative is capturing the demand that funds the descent. Learning rates are real; so is the volume that powers them.

3. Read the demand-side and supply-side tipping points separately, then together. Rogers governs the demand side: adoption tips when social proof crosses a threshold, not when the spec sheet wins. Abernathy and Utterback govern the supply side: the industry consolidates when a dominant design locks in. A true, hard-to-reverse tipping point is when both happen at once, each reinforcing the other. Diagnose them apart; respect them together. And treat subsidised volume with suspicion: a government-funded fleet can look like adoption, but it does not generate the peer-to-peer social proof that tips a Rogers curve. You cannot decree critical mass. Demand that vanishes when the subsidy stops was never adoption – it was a simulation of it.

4. Distinguish complicated from complex – and stop using Horizon 1 tools on Horizon 2/3 problems. Optimization, efficiency programs, and defending the installed base are the right tools for a complicated world where cause and effect are known. Aimed at a complex transition, they are worse than useless: they perfect the thing that is being made obsolete. The most optimized version of a dying design is still dying, only more efficiently.

5. Every new technology brings a new economics. Understand it, or optimize your way to irrelevance. This is the lesson beneath all the others, and it has played out for twenty-five years – in media, in retail, in AI, and now in mobility. Digital brought zero-marginal-cost scaling and network effects. Batteries bring a manufacturing learning curve and near-lossless energy delivery. Those who do not grasp the new economics will keep optimizing the old model with magnificent discipline – right up until it is too late. The perfect carriage was never the problem. Believing the race was about carriages was.

6. Remember that scale runs both ways. A learning curve that rewards the winner with falling costs punishes the loser with rising ones. When volume drains out of an incumbent system, its fixed-cost infrastructure – factories, refineries, networks – spreads over fewer units and the cost per unit climbs. The incumbent that looks unassailable at full volume can become structurally unviable well before the last customer leaves, because the economics break at the margin, not at zero. If your business rests on high throughput, the moment to ask what happens at lower throughput is before the volume starts to go.

7. Winning the model is not the same as owning it. Identifying the dominant business model is only half the task. The other half is owning the interdependent core components that make it defensible – what I call the UBM, the uniqueness of the business model. A borrowed core component is one block on someone else’s land, whether it is BMW’s fuel cell or Europe’s battery cell. The strategic question is never only “which architecture wins?” but “do I own the part of the winning architecture that carries the learning curve?” Europe answered the first question and neglected the second, and is on course to be a customer of the model rather than its author. Recognizing the winner too late, and building someone else’s advantage in the meantime, is its own kind of efficiency trap.

---

The powertrain race was never a referendum on which technology is cleverest. It was a competition between business models, decided by which value-creation architecture rides the steepest learning curve into the customer’s actual benefit – affordable mobility. Understand that, and you can stop arguing about molecules and start building, or funding, or joining the architecture that is going to win. Understand. Dream bigger. Act.

---

Annotated References

Abernathy, W. J., & Utterback, J. M. (1978). Patterns of industrial innovation. Technology Review, 80(7), 40-47. The founding statement of the dominant-design theory. Abernathy and Utterback showed that industries move from a fluid phase of competing product designs to a specific phase once a dominant design emerges, after which competition shifts from product innovation to process and cost, and the field of competitors thins in a shakeout. The core lens for seeing the powertrain contest as a consolidation event rather than an open technology comparison. See also Utterback’s Mastering the Dynamics of Innovation (1994) for the book-length treatment.

Arthur, W. B. (1994). Increasing Returns and Path Dependence in the Economy. University of Michigan Press. The formal economics of self-reinforcing advantage. Arthur showed that when a technology enjoys increasing returns to adoption – each adopter making the next more likely, through scale, learning, or network effects – markets can lock in on one option, and the lock-in may owe more to early momentum than to inherent superiority (his classic illustrations are QWERTY and VHS). This is the theoretical backbone for why the leading powertrain’s advantage compounds and why late challengers, however capable, struggle to reverse it. Overview.

BloombergNEF. (2026). Electric Vehicle Outlook 2026 and Behind the Scenes Take on Lithium-Ion Battery Prices. BNEF’s annual outlook and its battery-price survey. The outlook documents EV sales share, regional divergence, and charging-infrastructure growth; the price work establishes a battery learning rate near 18% per doubling of cumulative volume and pack prices around $100/kWh. Primary evidence for both the demand-side adoption data and the supply-side learning curve. Outlook; battery prices.

BMW Group / Driving Hydrogen / ArenaEV / pv magazine (2025-2026). Reporting on BMW’s iX5 Hydrogen programme: a pilot fleet of under 100 vehicles, series production planned for 2028 at deliberately low volumes, supported by €273 million in public funding (€191 million federal, €82 million Bavaria) under the IPCEI Hy2Move framework. The same reporting documents the fuel-cell sourcing history – first generation supplied entirely by Toyota, second-generation cells still sourced from Toyota, third-generation stack co-developed with Toyota to share R&D cost – the basis for the point that BMW does not own its core component. And BMW concedes that commercial success depends on hydrogen pump prices reaching parity with diesel, a learning-curve condition the programme’s volumes cannot meet. BMW pilot fleet; funding; fuel-cell sourcing; parity condition.

Carbon Direct. (2026). The sustainable aviation fuel cost premium is permanent – here’s what that means. An analysis of why SAF, and eFuels in particular, will not reach price parity with fossil jet fuel. Cites EASA’s 2024 production-cost estimate of ~€7,695 per tonne for power-to-liquid eFuels, roughly ten times conventional kerosene, and explains why neither airlines nor corporate buyers purchase SAF for its energy content. Key evidence that the eFuel cost gap is structural even in aviation, where no battery competes for volume. Link.

Christensen, C. M. (1997). The Innovator’s Dilemma. Harvard Business School Press; and Christensen, C. M., Kaufman, S. P., & Shih, W. C. (2008). Innovation Killers. Harvard Business Review, January 2008. The definitive account of why well-managed incumbents fail. Christensen’s paradox is twofold: good management practice – serving your best customers, optimizing existing processes – blinds established firms to disruptive shifts; and the disruptive technology, initially inferior on the dimensions incumbents’ customers value, improves along its own trajectory until it captures even the most demanding buyers. Innovation Killers adds the financial mechanism: judging new investments against the marginal cost of already-depreciated assets biases incumbents toward leveraging the old rather than building the new. Together, the dynamic behind the charging-speed argument and the intellectual neighbour of the efficiency-trap idea I apply throughout. Overview; Innovation Killers.

Clean Energy Wire (2026) and CleanTechnica (2025-2026). Reporting on the contraction of Europe’s hydrogen refuelling network for passenger cars. Clean Energy Wire documents Germany’s decline from over 90 to around 50 public stations; CleanTechnica details the closures in Germany and the complete dismantling of Austria’s public network, and works out the underlying throughput (roughly eight cars per station per day versus 200-300 for a petrol station). Evidence that hydrogen’s demand-side network effect for cars is running in reverse. Clean Energy Wire; CleanTechnica network math.

European battery supply chain: IEA (2025), Oxford Institute for Energy Studies (2025), Evolvance/IEA market data (2025-2026). Sources on where the battery core component is actually made. China controls more than 75% of global lithium-ion cell production and a larger share of cathode/anode materials; CATL holds close to 39.2% globally and over half in Europe. Europe’s home-grown champions (notably Northvolt) have largely failed, and Europe barely produces LFP cells – the dominant mainstream chemistry. The basis for the argument that Europe wins neither the losing nor the winning architecture at the level of the core component. IEA; Oxford Energy; market share data.

International Energy Agency (IEA). (2026). Ultra-fast charging batteries; with product data from Electrek and EV Charging Stations (2026). Evidence for the Christensen trajectory on charging speed. The IEA documents that battery charge rates have risen from roughly 2C on an average fast-charge session to 10C-12C in the early phase for the latest cells, enabling near-full charges in under ten minutes – comparable to refuelling. Product examples: BYD Blade 2.0 (10-70% in five minutes) and CATL third-generation Shenxing (10-98% in about six and a half minutes). Used to show the battery improving on the very dimension – charging time – that demanding customers cited as a reason not to switch. IEA; BYD; CATL.

Heavy-duty truck market and TCO studies: ICCT (China and Europe HDV market reports, 2025-2026), electrive (2026), IEA Global EV Outlook 2025 (heavy-duty chapter), Energy Innovation (2025), and peer-reviewed TCO analyses (ScienceDirect, 2025). The evidence base for the long-haul trucking boundary. The ICCT documents China’s zero-emission heavy-truck share reaching 25% in 2025 (over 450,000 units), nearly 30% in the heaviest tractor-trailer segment, and a market already consolidating around five leading makers (about 61% share, including SANY) – a dominant-design signal. Europe sits far earlier on the curve (~4.5% share) but grew about 60% year-on-year as CO₂ standards took effect. The IEA and Energy Innovation find battery-electric trucks reaching or beating diesel on total cost of ownership by around 2030, with fuel-cell trucks remaining more expensive. A dissenting peer-reviewed strand still finds diesel cost-optimal across many ranges and weights, partly because long-haul diesel engines run near 45% efficiency (versus ~35% for passenger petrol engines), narrowing the electric advantage. Cited to keep the boundary honest: the segment is moving toward batteries, already tipped in China, later elsewhere, with real open questions. ICCT China; ICCT Europe; electrive; IEA heavy-duty; Energy Innovation; dissenting TCO study.

International Council on Clean Transportation (ICCT). (2021). E-fuels won’t save the internal combustion engine; and (2025). Why and how to bring down the cost of SAF. Two ICCT analyses. The first is a technical accounting of the conversion losses in synthetic fuels: roughly half the input electricity is lost simply converting power to liquid, before the ~35%-efficient engine even burns it – the thermodynamic floor beneath the eFuel cost position that no learning curve can lift. The second calculates a 2.1-10.6× cost premium for sustainable aviation fuels over fossil jet fuel and projects that eFuels remain far above parity even after 2030. E-fuels won’t save the ICE; SAF cost.

International Energy Agency (IEA). (2026). Global EV Outlook 2026. The most comprehensive annual dataset on electric mobility. Sources here for 2025 global EV sales exceeding 20 million and ~25% sales share, China’s ~48% NEV share, battery deployment, and the raw-material price movements (notably the early-2026 lithium spike) that bound the claim. Trends chapter; batteries chapter.

Liebreich, M. (2023). Clean Hydrogen Ladder, Version 5.0. Liebreich Associates (CC-BY 4.0; concept credit: Adrian Hiel, Energy Cities). The most useful framework for keeping the hydrogen debate honest in both directions. Ranks use cases from A (no alternative: fertiliser, hydrocracking, desulphurisation, methanol) down to G, the “Row of Doom”, weighing cost, efficiency, safety, critical-mineral availability and whether a cheaper alternative exists. Passenger cars sit at the bottom. Cited here for the distinction the powertrain debate keeps collapsing: hydrogen is indispensable in industry and uncompetitive in the car. Link.

International Energy Agency. (2026). Global Hydrogen Review 2026 (Production chapter). The authoritative annual stock-take, and the basis for the second warning in this piece. Documents that low-emissions hydrogen production reached about one million tonnes in 2025 – the first time above 1% of total output – while new final investment decisions fell below 0.8 Mt a year, the pipeline to 2030 shrank by 10 Mt to 27 Mt, and over 100 GW of announced electrolysis capacity risks never operating before 2030. The IEA notes electrolyser manufacturing entering a consolidation phase because the market is developing too slowly. Cited for the point that even in the applications where hydrogen is unavoidable, it sits near the top of its own learning curve. Link.

IRENA. (2020). Green Hydrogen Cost Reduction: Scaling up Electrolysers to Meet the 1.5°C Climate Goal. The reference study on hydrogen’s cost trajectory. IRENA derives electrolyser learning rates in the 16-21% range and quantifies the enormous cumulative investment required to bring green hydrogen to cost parity – evidence that hydrogen has a learning curve, but starts from a high cost position and needs volume it is not capturing. Link.

Oliver Wyman (2022), Semiconductor Shortage in the Auto Industry; EE Times (2026), Automakers Face Memory Crunch as AI Strains Chip Supply; and TSMC customer data (CNBC, 2026; industry analyses, 2025-2026). Evidence for the reversal of supplier bargaining power. Oliver Wyman documents that automotive accounts for only around 8% of global semiconductor sales and that carmakers “do not play a dominant role in the semiconductor industry.” EE Times quotes a Gartner analyst noting that at roughly 10% of the market, automotive is “by no means a high priority” for chipmakers. The TSMC figures make it concrete: Nvidia is the foundry’s largest customer at roughly 22% of revenue, Apple second, the top five accounting for an estimated 65-70% of the total, while the entire global automotive sector represents about 7% of revenue. Cited for the point that ten million vehicles conferred power only while the car industry was the largest buyer in its component markets. Oliver Wyman; EE Times; CNBC on Nvidia overtaking Apple; TSMC segments.

Our World in Data. (2026). Battery costs have declined by 99% in the last three decades. The clearest public presentation of the battery learning curve. Documents that since 1998, each doubling of cumulative lithium-ion production has cut prices by about 19%, sustained across a 27-million-fold rise in cumulative output – the empirical heart of the supply-side argument. Link.

Ziegler, M. S., Song, J., & Trancik, J. E. (2021). Determinants of lithium-ion battery technology cost decline. Energy & Environmental Science, 14, 6074. MIT research disentangling why lithium-ion costs fell, rather than merely observing that they did. Documents a roughly 97% price decline across three decades of commercialization and quantifies the contributing mechanisms. Cited here for the point that the battery’s descent down its learning curve began in consumer electronics, long before automotive volume existed. See also Ziegler & Trancik (2021), Re-examining rates of lithium-ion battery technology improvement and cost decline, which distinguishes cells designed for consumer electronics from those built for other applications. Determinants; Re-examining.

Rogers, E. M. (2003). Diffusion of Innovations (5th ed.). Free Press. (Original work published 1962). The foundational theory of how innovations spread through social systems. Rogers established the adopter categories (innovators, early adopters, early and late majority, laggards) and the central role of interpersonal communication and social proof in driving adoption past a threshold. The basis for reading EV demand as a Rogers S-curve that has crossed into the majority. Overview.

Saudi Aramco financials (Investing.com, 2026) and IMF fiscal break-even estimates (via OilPrice, 2025). The cost structure of the lowest-cost oil producer. Aramco reports lifting costs around $3.50 per barrel (CEO Nasser cites ~$2 extraction cost) against a 2025 realized price near $69 – a wellhead contribution margin above $65 per barrel. Yet Saudi Arabia’s fiscal break-even, per the IMF, sits around $91-96. The gap illustrates the argument that the fossil chain is propped up by extraordinary upstream margins, while the rentier economy on top of it assumes both high prices and high volumes – the vulnerability behind the reverse-scale effect. Aramco costs; fiscal break-even.

SNE Research / electrive / Driving Hydrogen (2025-2026). Market data on global fuel-cell passenger-car sales. Establishes the absolute figures used here: 16,011 hydrogen cars sold worldwide in 2025, a peak of 20,704 in 2022 followed by annual decline, 485 units in Europe in the first half of 2025, and the ~70% drop in German registrations in 2023. The order-of-magnitude gap to 20-plus million EVs that grounds the “Rogers in reverse” reading. Global 2025 figure; H1 2025 decline.

Stähler, P. (2001). Geschäftsmodelle in der digitalen Ökonomie. Josef Eul Verlag. My dissertation. Among the first works to establish the business model as a distinct unit of analysis, and the source of two ideas used here: the formal derivation of the word-of-mouth tipping point, and the prediction that competition among business models produces dominant designs and temporary oligopolies driven by demand-side network effects and supply-side scale. ResearchGate.

Stähler, P. (2021). Das Richtige gründen: Werkzeugkasten für Unternehmer (5th ed.). Murmann. The source of the four-element business-model framework used as the diagnostic grid throughout this post – value proposition, value-creation architecture, revenue model, and company spirit – and of the customer-benefit discipline (the customer buys the benefit, not the product).

World Economic Forum. (2025). The cost of sustainable aviation fuel: Can the industry clear this key hurdle? A concise statement of the SAF scaling problem, framing it explicitly as a chicken-and-egg dynamic: airlines signal demand, but supply remains limited and prices are projected to stay two to three times higher than fossil jet fuel through 2030. Support for the argument that demand signals alone do not walk a high-cost fuel down its learning curve. Link.

Wright, T. P. (1936). Factors affecting the cost of airplanes. Journal of the Aeronautical Sciences, 3(4), 122-128. The original learning-curve paper. Wright observed that airplane production costs fell by a consistent percentage with each doubling of cumulative output – the empirical regularity, now called Wright’s Law, that underlies the battery cost decline and the entire supply-side argument of this post.