Cars used to be mechanical things with a few chips bolted on. Now it’s flipped — a modern vehicle ships with more code than a fighter jet, and the hardware is almost an afterthought. That’s the software-defined vehicle, or SDV: a car where features and even driving feel come from updates, not a dealership visit. A lot of automakers are behind on this. Here’s why and what’s actually being built right now.

The platform problem nobody likes admitting

Hyundai didn’t want to find this out the hard way. Working with DXC on a next-gen infotainment platform meant building a production-ready, Android Auto-based cockpit system from scratch instead of stitching together whatever the Tier 1 suppliers handed over. That’s basically the whole argument for platforms over point solutions right now. DXC built its own answer to the integration mess, an SDV platform meant to pull cockpit, cluster, autonomous drive, and real-time domains into one stack instead of a dozen disconnected ones.

Why the old model stopped working

Here’s the part nobody likes saying out loud: the traditional supply chain was built for hardware, not software. A typical car carries 100 to 150 electronic control units, each running its own firmware, sourced from a different vendor, talking over CAN bus protocols designed in the 1980s. Want one new feature? You’re touching five ECUs from five suppliers, each on its own release schedule. Good luck.

Tesla solved this years ago by collapsing dozens of ECUs into a handful of domain controllers. Everybody else is still playing catch-up. Volkswagen’s Cariad division burned through billions trying to build a unified stack in-house and is still wrestling with delays on the Porsche Macan EV and Audi Q6 e-tron. Rivian, meanwhile, licensed its software stack to Volkswagen for a reported $5.8 billion — a pretty blunt admission that this stuff is brutally hard to build from zero, even with deep pockets.

So what changes with an SDV architecture?

Centralized compute swaps dozens of scattered ECUs for a handful of powerful domain or zonal controllers
Over-the-air updates push new features, fixes, even performance tweaks, no service visit needed
Decoupled hardware and software lifecycles, so the chip you picked five years ago doesn’t dictate what ships today
Standardized middleware lets different teams, and different suppliers, build on one shared foundation instead of rewiring everything every time

Sounds logical. The hard part is execution, not the concept.

The compute architecture shift

Zonal architecture gets thrown around a lot these days, so it’s worth unpacking. Rather than running a separate computer for every function (one for seat memory, one for mirrors, one for infotainment) manufacturers group electronics by physical location (front, rear, left, right) and run them off a shared zonal controller. NVIDIA’s DRIVE Thor, built on the Blackwell architecture, is pitched as the brain for exactly this setup, handling infotainment rendering and autonomous inference on one chip. Qualcomm’s Snapdragon Digital Chassis is chasing the same goal from another direction, already running cockpit systems in BMW, Mercedes and Honda vehicles.

What’s getting tested on the show floor right now

CES 2026 in Las Vegas was basically a software trade show wearing a car costume. Sony Honda Mobility showed off updated Afeela software with deeper ties to Epic Games’ Unreal Engine for cabin visuals — yes, the engine behind Fortnite is now rendering dashboards. BYD demonstrated its “God’s Eye” driver assistance suite running across its full lineup, which is pressuring competitors to stop treating ADAS as a premium add-on. DXC used the show to introduce its AMBER automotive software platform, built for exactly the integration headaches above.

A few threads worth pulling on:

AI inside the cabin. Voice assistants are gaining actual reasoning instead of canned responses — Mercedes-Benz’s MBUX now runs on Google’s Gemini models for conversational navigation and vehicle queries.
Digital twins for validation. Before a single line of production code ships, companies like Ansys and Siemens run entire vehicle subsystems as virtual models, catching bugs that would’ve meant a recall six years ago.
Software factories. CI/CD pipelines, borrowed straight from tech, are getting bolted onto automotive development so updates ship in days instead of years.

Who’s testing what, and where

Mobileye keeps racking up robotaxi pilots, lately expanding with Volkswagen Group and Uber for autonomous rides across Europe. Waymo, still the most mature player by sheer mileage, passed 100 million driverless miles on public roads. On the more experimental end, Polestar has been testing over-the-air torque vectoring — the literal handling of the car changes after a software push, no mechanical work required.

Why suppliers are racing to lock in OEM partnerships now

There’s a quieter story running underneath all the CES headlines: the window for picking a software partner is closing fast, and everyone in the supply chain knows it. Once an automaker commits to a middleware stack or a compute platform, ripping it out three years later costs nine figures and a generation of engineering headcount. That’s exactly why Qualcomm, NVIDIA and a wave of smaller players are signing multi-year design wins before a single production vehicle rolls off the line.

Mercedes locked in Qualcomm’s Snapdragon Digital Chassis across its entire lineup through the next decade. Foxconn, of all companies, is positioning itself as a contract manufacturer for SDV hardware the same way it builds iPhones, with its MIH open platform aimed at smaller automakers who can’t afford to build compute architecture from scratch. Even Renault and Google deepened their Android Automotive partnership specifically to avoid the fate Cariad ran into — building everything in-house and missing every deadline along the way.

What does this mean for buyers of these platforms? Mostly that the decision isn’t really about today’s feature list. It’s about who’s still going to be shipping updates, patches and new hardware support five years out. A platform vendor who disappears, gets acquired, or quietly stops investing leaves an automaker stuck maintaining dead software on millions of cars already on the road. Worth asking before signing anything: who’s actually funding the roadmap, and for how long?

Where the actual money and risk sit

Building an SDV stack from scratch is expensive, slow, and full of dead ends most legacy automakers can’t afford to hit twice. That’s the gap platforms exist to close. DXC’s AMBER leans on a modular foundation, pre-integrated cockpit, cluster and middleware components, instead of asking a manufacturer to build every layer in-house. The pitch: cut SDV development time by roughly half, trim infotainment development costs by close to a third, according to the company’s own published numbers, while keeping the architecture open enough to avoid vendor lock-in down the road.

That last part matters more than it sounds. Plenty of automakers got burned in the early 2010s by infotainment suppliers that locked them into proprietary stacks that aged badly. Nobody wants a repeat of that with autonomous driving software baked in this time.

A few risks worth watching:

Cybersecurity exposure grows with every connected feature — UNECE’s WP.29 regulation now requires a formal cybersecurity management system for type approval in major markets
Talent is scarce; automakers are competing directly with Silicon Valley for embedded and cloud engineers, not exactly home turf for them
Regulatory fragmentation between the EU, US and China means a feature cleared in one market can stall for months somewhere else

What it means for the next few years

Companies that figure out software velocity (shipping features fast, safely, without breaking what already works) are going to pull ahead the same way smartphone makers separated from feature phone makers fifteen years ago. The ones still treating software as a bolt-on keep bleeding money on integration work that platforms like AMBER are explicitly built to eliminate.

Will every automaker get there on its own timeline? Doubtful. Some will buy their way in through licensing deals like Rivian’s. Some will lean on platform partners. A few will stumble badly before figuring it out — Cariad’s struggles prove that throwing billions at the problem doesn’t guarantee a working product. But the direction is set. The car is a computer with wheels now, and the manufacturers treating it that way are the ones worth watching.