Evolution of EE Architectures in Cars to 2030+

Siemens, Markets and Markets

Picture two SUVs rolling off two different assembly lines on the same morning in 2028. Same segment, same sticker price, near-identical spec sheets. Three years later, one of them has quietly taught itself to drive hands-free on the highway, parks itself in your garage, and carries features its owner never paid for on the day of purchase. The other is exactly the car it was the morning it was sold, no smarter, no faster to react, frozen in time. The two vehicles don’t differ in horsepower, badge, or even price. They differ in something almost nobody shopping for a car has ever heard of: their E/E architecture, the electrical and electronic nervous system that decides whether a vehicle can keep evolving after it leaves the factory, or whether it is already as good as it will ever be.

Most people will never use the term. Yet the way a car’s electronics are wired together has quietly become one of the most consequential strategic decisions an automaker makes, more consequential, arguably, than the engine, the battery, or the badge on the hood. E/E architecture is simply the blueprint for how all the computers, sensors, and wiring in a vehicle are organized and talk to each other. For a century it was an afterthought, a cost to be minimized. Today it is the foundation on which a carmaker's entire future business model is being built.

So what exactly is changing? To put it simply, the car is going from a collection of many small brains to a single large one. For most of automotive history, a vehicle’s electronics grew like a city without a planner. Every new feature, anti-lock brakes, a parking sensor, a heated seat, lane-keep assist, arrived as its own dedicated computer, an Electronic Control Unit, or ECU, usually supplied and coded by a different vendor. By the 2010s a typical car carried 100 to 150 of these ECUs, stitched together by four to six kilometers of copper wiring weighing 40 to 60 kilograms, the third-heaviest component in many vehicles. It was expensive, heavy, and, critically, impossible to update as a whole. Each computer was a sealed black box. The carmaker often didn't even own the software running inside it.

That model is now collapsing, and fast. The industry is migrating along a clear path: from distributed ECUs, to function-based "domain" controllers, to location-based "zonal" controllers, and finally to one or two central compute units that act as the car’s brain. The numbers are striking. ECU counts are dropping from over 100 to as few as five to ten. Wiring harnesses are shrinking 30 to 50 percent. A modern zonal design can strip 20 kilograms and a couple of miles of wire out of a single vehicle. Tesla, with the EtherLoop ring it is rolling out on its next hardware generation, claims to cut wiring by as much as 68 percent.

Why is this happening now, after decades of inertia? A few forces have arrived at once. Cars have become software-defined (shifting to AI-Defined), and updating dozens of siloed computers over the air is effectively impossible, you need one consolidated brain. Advanced driver assistance has gone from needing around 10 trillion operations per second a few years ago to needing more than 2,000 today, a level of real-time sensor fusion that fragmented little computers simply cannot deliver. Electrification demands tighter management of 800-volt, 48-volt, and 12-volt systems in one place. And then there is the reason that matters most in the boardroom.

This is the part that is widely misunderstood. The rewiring of the car is not, at its heart, an engineering upgrade. It is a business-model transformation. For a hundred years, a carmaker sold you a vehicle once and made its money at the moment of sale. The new architecture turns that logic inside out. When hardware is decoupled from software and the whole car can be updated over the air, the vehicle stops being a product you sell once and becomes a platform you sell, and re-sell, for fifteen to twenty years. Tesla charges roughly $99 a month for Full Self-Driving. BMW has sold heated-seat subscriptions. GM’s Super Cruise runs around $25 a month. Rivian prices its autonomy package near $50 a month. None of this recurring revenue is possible on the old distributed architecture, because you cannot remotely activate or improve a feature that lives inside a sealed supplier box. The E/E architecture, in other words, has quietly migrated from being a cost center to being the single most important enabler of an automaker's future earnings. My team at Markets and Markets believes, car companies can generate upto $1600 per vehicle per annum with this architecture in 2035.

Which raises the question the entire industry is now circling: if software and compute are the prize, who actually owns them? Here the field splits sharply, and not along the lines you might expect. Two companies, Tesla and Rivian, have achieved what the rest of the industry has not: full "silicon-to-cloud" ownership at production scale. They design their own chips, write their own operating systems, build their own middleware, and run their own clouds, with no Tier-1 supplier sitting in the revenue path. Tesla’s coming AI5 chip targets around 2,500 trillion operations per second and is being dual-sourced from TSMC and Samsung, with Tesla as the direct foundry customer, an arrangement that gives it supply-chain immunity no traditional carmaker enjoys. Rivian's in-house RAP1 silicon and its Safe ARTUS operating system make it the only mass-market automaker besides Tesla to own the entire stack. That ownership is not a feature. It is a moat that competitors, by most estimates, cannot close within five years.

Everyone else is, to varying degrees, renting their brain. GM, Ford, BMW, Mercedes-Benz, Volkswagen, and Toyota still depend on Qualcomm, NVIDIA, and Mobileye for the compute at the center of their vehicles. Some are racing to change this, and the scars are real. Ford reportedly spent heavily on its ambitious "FNV4" zonal architecture without shipping a single production model, then retreated to a more modest "FNV3.X" design that delivers most of the commercial benefit at a fraction of the cost. GM, by contrast, has placed a decisive bet, building a centralized platform on NVIDIA’s Drive Thor that collapses more than 100 ECUs into roughly five compute nodes and debuts on the Cadillac Escalade IQ in 2028 before cascading down to mass-market models by 2032. The fact that these are excellent chips from excellent suppliers is, in a sense, beside the point. Whoever controls the silicon and the software controls the post-sale revenue. Everyone else is a tenant.

And the dependency runs deeper still, because every one of these companies sits atop a supplier base that is now staring at structural decline. As ECU counts fall by close to 90 percent and harnesses shrink by up to half, the traditional wiring-harness giants, Yazaki, Sumitomo, Leoni, Lear, face a direct volume threat, and pure ECU suppliers face consolidation. The winners among the Tier-1s will be those who climb the value chain into Ethernet networking, zone-controller modules, and integrated software stacks; companies like Aptiv, Bosch, and Continental are already trying to reposition from hardware vendors into system integrators. The window to make that pivot is narrow, roughly 2025 to 2028, after which platform-sourcing decisions lock shut for the better part of a decade.

There is also a fascinating strategic wrinkle that hints at where this all goes next. If the architecture is the prize, it can also be sold. Rivian’s roughly $5.8 billion joint venture with Volkswagen Group, which licenses its zonal architecture and Safe ARTUS software across VW's brands, sets a precedent: the E/E architecture itself is becoming a licensable asset. For carmakers that cannot afford to build a full software-defined platform from scratch, and most cannot, given the cost and time-to-market risk, licensing someone else's may become the only rational path. The car company of the future may run on another car company's operating system, the way much of the computing world runs on Windows or Android.

So what is the strategy for those still holding legacy architectures? Honestly, there are fewer good options than Markets and Markets suggest. Building a proprietary chip and OS is enormously expensive and slow. Licensing cedes a measure of independence. Doing nothing is fatal: by 2030, essentially every OEM will have moved to a zonal or centralized design, pushed along not only by economics but by regulation, UN rules R155 and R156, on cybersecurity and over-the-air updates, make the old sprawl of 100-plus attack surfaces structurally expensive to defend. The transition is no longer a question of *if* or even *when*. It is a question of *who owns what* on the other side. And on that question, the gap between the companies that control their own silicon and software and the companies that lease it is the gap that will separate the winners from the also-rans.

The future of the car is intelligent, connected, and continuously improving. But that future belongs only to the companies that own the nervous system underneath it, the wiring, the compute, the code. The rest will keep building beautiful machines that go quietly out of date the moment they leave the lot. The road is forking right now. And it does not wait for anyone to make up their mind.