Porsche Didn't Lose to the Battery: The Category Transition AI Adopters Keep Missing

Early on the technology, late on the evaluation basis — why “legacy product + new tech” keeps losing, and the cockpit rule AI teams should steal.

Scott Farrell · LeverageAI · August 2026

Porsche’s own investor materials for the 2025 financial year are blunt. Group sales revenue fell to €36.27 billion (from €40.08 billion in 2024). Group operating profit fell from €5.64 billion to €413 million. The Group operating return on sales fell from 14.1 per cent to 1.1 per cent. Extraordinary expenses of roughly €3.9 billion — product-strategy realignment and rescaling, battery-related costs, and US tariffs among them — sit inside that collapse.

Those are extraordinary numbers for a brand built on premium margins. They are also not the end of the company. The same first-party materials stress a strong financial basis, high net liquidity, and a healthy balance sheet. In the first quarter of 2026, Porsche’s own figures show a group operating return on sales of 7.1 per cent — recovered into the upper end of its full-year forecast range, not a death spiral. By the first half of 2026, Porsche’s own half-year release puts Group return on sales at 7.8 per cent and reaffirms the full-year outlook, with automotive net liquidity of €7.3 billion.

So the accurate sentence is this: Porsche suffered an earnings collapse and a strategic identity crisis, and is now attempting a major turnaround. Keep that sentence in your head. Everything that follows is about the structure of the mistake and the shape of the correction — not a eulogy.

This piece is an extender of Stop Automating, Start Replacing — the horse/car taxonomy for AI projects: do not optimise the inherited process; replace the machine when the category has moved. I will not re-teach that taxonomy here. What this piece adds is the most vivid current public parable for that framework — plus a subtler lesson the slogan often misses: you can adopt the new technology early and still miss the category transition entirely.

The mis-read: early on propulsion, late on the evaluation basis

The shallow story about Porsche is easy to tell: they charged into electric vehicles ahead of customer readiness, Chinese software-defined competitors undercut them on price and features, tariffs and policy headwinds made everything worse, and the margin structure cracked.

Parts of that story are real. First-half 2026 deliveries, per Porsche’s own newsroom, fell 16 per cent worldwide (122,306 vehicles against 146,391 a year earlier). China fell 32 per cent. Independent wire coverage reported the same delivery shape — pressure in China and the expiration of US tax incentives among the cited factors.

But the deepest problem was not simply “they chose EVs too early.”

It is closer to this:

Porsche treated the propulsion transition as the main variable while the basis on which many customers — especially Chinese buyers — evaluated a premium car was also changing.

That re-basing is not about horsepower. Once the powertrain is electric, raw performance stops being a scarce differentiator in the way it was when a combustion sports car had to earn its lag and its soundtrack. “All EVs have performance” is not a slogan against engineering excellence. It is a market observation about scarcity: if torque is abundant, the buyer’s scarce attention moves to software responsiveness, infotainment quality, local digital ecosystems, development speed, integrated features, and price/performance that feels contemporary rather than archaeological.

Porsche’s own actions admit this. In November 2025 it opened its first integrated overseas R&D hub in Shanghai under an “In China, for China” framing — explicitly to connect German engineering with China’s digital ecosystem, cut development cycles from years to months, and ship a China-exclusive next-generation infotainment system. Independent coverage earlier the same year had already reported management signalling a China-exclusive infotainment solution for 2026.

That is not a press-release flourish. It is an institutional admission that German vehicle engineering excellence alone was no longer sufficient product fit in a market where the category’s evaluation basis had shifted toward software and integrated digital experience.

The mis-read, in other words, is structural:

• Technology variable: move the powertrain from combustion to battery.
• Category variable (missed or late): what buyers use as the basis of judgment — interface coherence, software currency, local ecosystem fit, simplicity of use — while complexity stays under the skin.

You can win the first and still lose the second. That is the lesson AI adopters keep missing when they celebrate “we already have copilots in the process” as if the technology variable were the strategy.

Archaeology in the cockpit

For decades, complicated deterministic cars could be sold as feature catalogues. Buttons, dials, mode switches, option packages, the visible wiring of capability — mechanical and electronic complexity was not only tolerated; it was often a virtue. You could feel the archaeology of automotive history in the cabin. Luxury European brands were especially good at this: each control was a small proof that something sophisticated was happening.

Software-defined vehicles invert the aesthetic of that virtue.

The underlying machine is not simpler. If anything, the stack is denser: sensors, over-the-air update paths, inference, connectivity, local services. What changes is the interface contract. Complexity is hidden behind a fitted, coherent interaction model. The driver is not invited to navigate the history of the platform. They are invited to drive — and to use a small number of surfaces that feel current.

That is the “archaeology in the cockpit” rule, stated as a product law:

Feature-catalogue thinking fails the first half of that rule. Black-box “just trust the AI” thinking fails the second. Both are category errors about what the interface is for.

Chinese EV makers, as a category, forced this contrast into the open: cars bought less as catalogues of switches and more as coherent digital experiences on wheels. You do not need a model-by-model industry survey to see the structural point. When competitors re-base the product around software-defined simplicity, an incumbent that merely electrifies the old cockpit is answering a question the market has partially stopped asking.

EV is not “a battery and an electric motor for a legacy car.” That sentence is the whole case in miniature.

The AI mapping: the same weak transition, different industry

Now map the structure without stretching the metaphor past what it can carry.

A software-defined vehicle does not expose the driver to the full archaeology of automotive engineering merely because the underlying vehicle is complex. The complexity is real. The interaction model is fitted.

Likewise, an AI-native workflow should not force an employee or a customer to navigate:

• the warehouse,
• the document repository,
• five applications,
• retrieval tools,
• raw prompts,
• agent traces,
• and a separate approval system that feels bolted on after the fact.

That is archaeology in the cockpit — enterprise edition. It is the same category error as selling a thousand switches because the wiring harness is complicated.

The interface should organise around a different primitive. In our own work we call this decision-navigation UI: the unit of interaction shifts from “what do you want to see?” (record navigation / warehouse spelunking) to “what do you want to decide?” (proposal, evidence, disposition).

Concretely, the surface should answer four questions in one review unit:

That is the cockpit rule applied to organisational cognition. Hide the warehouse, the five applications, the agent’s intermediate thrashing, and the prompt archaeology. Do not hide the basis of judgment. Progressive disclosure can still offer full records on demand — Layer 3 for the edge case — but the default unit is a decision package, not a tour of internal systems.

If your AI project still requires the user to be the join algorithm between systems — assembling context by hand, then pasting it into a chat window, then re-entering the result into three other tools — you have built legacy workflow + AI assistant. You have electrified the cockpit without redesigning it.

That is why adding technology keeps losing to natively redesigned competitors. The competitor is not merely “more AI.” The competitor is answering a different evaluation basis: coherence, speed of fit, fewer handoffs, a surface organised around outcomes rather than internal archaeology.

The recovery shape has to match the doctrine

If the mistake is structural, the recovery cannot be “more battery” or “more chatbot.” It has to match the same doctrine the parable implies.

Porsche’s current public strategy language — Strategy 2035, as outlined at the 2026 Annual General Meeting — is striking in this light. The three pillars Porsche itself emphasises are, in substance:

• Brand and customer: concentrate on sports-car DNA; remain a brand for people who consciously want to drive, especially in an automated world; pursue value, desirability and profitability rather than volume for its own sake.
• Products and technology: reduce the number of model variants; focus more sharply; invest across combustion, hybrid and electric; make EVs that are distinctively Porsche rather than generic.
• Company and operations: streamline the organisation; reduce complexity; align structure to core business.

That is not “become a cheaper Chinese EV maker.” It is closer to a recovery doctrine that maps cleanly onto AI workflow redesign:

The 911 figure matters here precisely because it prevents a eulogy reading. Porsche’s own H1 2026 delivery release states that 911 deliveries rose 19 per cent while overall deliveries fell 16 per cent. That does not prove the whole turnaround. It supports a narrower, usable claim: the strongest differentiated product can remain valuable even as the surrounding portfolio and category transition are painful. Harvest the loyal value pool. Do not pretend every SKU is equally strategic.

Q1 and H1 margin recovery into the mid-to-high single digits, on Porsche’s own figures, is the same kind of signal for the corporate story: severe deterioration, then deliberate recalibration — not insolvency theatre.

What to ask before you fund the next “technology” move

The reader question this piece is built to answer is practical:

Why does adding new technology to an existing product or process keep losing to natively redesigned competitors?

Because the competitor often changed the evaluation basis — not merely the component. You optimised the technology variable. They redesigned the product around how buyers now judge the category. You shipped legacy + battery. They shipped a different machine with a fitted interface. Your demo works. Their product fits.

Before the technology question, ask the category-transition question:

1. What is the buyer’s (or user’s) new evaluation basis? Not last decade’s. Not your internal process map’s. The basis they use when a native alternative is available.
2. If we only changed the technology and left the interface and architecture alone, would we still look contemporary? If the honest answer is no, you are planning legacy + assistant.
3. Does our interface hide machinery or hide judgment? Apply the cockpit rule. Warehouse spelunking and agent-trace tourism fail the test. Decision packages with evidence and explicit consequences pass it.
4. Where should we be native, and where should we harvest loyalty? Not everywhere at once. Reduce false variants. Be distinctively yourself only where that distinctiveness is real.
5. Are we celebrating early technology adoption as if it were category leadership? Early batteries without software-fit still lose. Early copilots without workflow redesign still lose.

If you need a one-line lint for project proposals, use this:

Sibling work already covers adjacent discipline: fixed-price certainty before implementation; when a knowledge system kills bad projects better than it invents good ones; how decks and documents become software. Use those when the question is commercial packaging or pre-mortem lint — not as substitutes for the category question this piece owns.

Close

Porsche did not “lose to the battery.” Batteries are a component. The harder loss is missing a category transition: buyers re-basing what “good” means while the incumbent answers with a technology swap inside an inherited product logic.

AI adopters are running the same experiment every quarter. They bolt an assistant onto the inherited workflow, measure step-speed, declare early adoption, and then lose to someone who redesigned the work so the human never had to tour the warehouse. The demo looks modern. The category judgment is still archaeological.

The takeaway is not “avoid technology.” It is sequence and design:

• Ask what the new evaluation basis is before you ask which model or motor to buy.
• Redesign the surface of judgment — decision, evidence, uncertainty, consequence — rather than electrifying the old cockpit.
• Recover like the doctrine: fewer variants, loyal value pool, digital baseline, native only where you can be yourself.

Hide the machinery. Do not hide the basis of judgment. That is the cockpit rule — for cars, and for AI.

If this reframes a project you are about to fund, the useful next move is not another tool bake-off. It is a one-page category-transition note: the old evaluation basis, the new one, and whether your current plan is legacy-plus-tech or a native redesign. Start there. The technology choice gets easier after that.