Interview: How Volvo constructed software program for a two-and-a-half-tonne shifting object
Anders Bell factors to his gray hair and laughs. “Three years in the past, it was nonetheless blond and curly,” says Volvo’s chief engineering and expertise officer. The comment is greater than self-deprecating. It captures what Volvo has been by: 5 years of constructing a software-defined car (SDV) from scratch, as a conventional carmaker, with no blueprint to comply with and no provider to name for the exhausting elements.
The EX60 (pictured above) is a mid-size all-electric SUV, unveiled in Stockholm in January 2026 and constructed on SPA3, Volvo’s new electrical car structure. In April, the primary buyer manufacturing automobile rolled off the road on the Torslanda plant in Gothenburg. It’s the first constructed solely on HuginCore, the corporate’s personal computing platform.
S&P International Mobility charges Volvo at Stage 5, the very best class of software-defined car functionality, and it’s the solely producer on the planet to succeed in that stage. Bell describes the journey with a vivid metaphor: “Take a chunk of charcoal, strain take a look at it sufficient till it turns into a diamond. Now we’re sprucing the diamond.”
On the coronary heart of HuginCore is a single software program grasp: one codebase, configured per car, shared throughout each mannequin on the platform. A developer engaged on a door locking operate doesn’t launch code to the EX60. The code goes to the grasp. From there, it’s configured for each automobile.
The EX60 is the third Volvo to run on that grasp, following the EX90 and ES90. Every one hardened the codebase additional. “That is principally what Apple does. That is what Microsoft does. If you wish to be nice at software program, you must solely have one.”
Laborious classes realized from real-world buyer expertise
The EX90, the primary automobile on the platform, had a tough introduction. Issues that automated testing had not caught emerged in the true world, and clients skilled them straight. “It’s been a tough and painful journey,” Bell says. “I don’t thoughts if it’s exhausting and painful inside the corporate. It’s very unlucky when it spills over to clients.”
Creating its personal computing platform additionally modified how Volvo works with suppliers. Conventional tier-one suppliers ship full digital management items (ECUs), {hardware} and software program as a single bundle. With HuginCore, Volvo developed its personal zone controllers and integrates provider software program itself.
“We don’t simply take an ECU from Bosch that’s prepared,” says Bell. “We now have to work with them, combine their software program on our zone controllers, so their system works and integrates by the automobile.”
“I might by no means dare to press the button on 2.2 million vehicles and simply go. A failed replace that may merely be retried: superb. An replace which means you may now not begin the automobile: catastrophic”
Anders Bell, Volvo
That shift provides Volvo management over the combination layer, and that management has penalties that run deeper than structure diagrams counsel.
Bell illustrates it with a concrete instance. At excessive chilly, round -25°C, the battery wants safety from freezing. The usual answer is a separate electrical heater costing round €150. Volvo requested a distinct query: might the electrical motor produce warmth deliberately, by operating barely much less effectively in excessive chilly?
The reply was sure. The separate heater was eradicated. “We calculated that for €8 of further elements within the motor, we might take away a €150 half elsewhere,” says Bell.
That type of cross-system optimisation solely turns into potential when totally different engineering disciplines sit on the identical desk and deal with the car as one built-in system reasonably than a group of separate modules. Within the outdated organisational construction, Bell says, that dialog would by no means have occurred.
Product releases should be nothing lower than superior
The sensible attain of the one software program grasp turns into clear in deployment. Volvo is presently rolling out a one human-machine interface replace to 2.2 million vehicles constructed way back to 2020, bringing a standardised interface and Google Gemini conversational AI to automobiles that left the manufacturing unit earlier than massive language fashions (LLMs) existed.
“We had no concept in 2020 after we constructed these vehicles that we’d, in 2026, roll out conversational AI on the identical automobile, on the identical {hardware},” says Bell. “We didn’t know what an LLM was in 2020.”
A software-defined product is rarely completed. There’s all the time extra to combine, extra performance so as to add. However there’s a second whenever you resolve it’s ok to launch. Bell’s criterion is pragmatic: “Once you launch it, it should be freaking superior.” And at that second, it’s.
However growth continues, and what feels groundbreaking right this moment will really feel dated in 5 years. That, Bell argues, just isn’t an issue. It’s the level. The purpose is not only to fulfill clients in the mean time of buy, however to maintain stunning them afterwards. The conversational AI now rolling out to vehicles in-built 2020 is precisely that.
The engineering behind these deployments is equally instructive. Volvo builds 20 full software program variations per day. Each 4 hours, the very best candidate goes by full automated car testing on rigs, across the clock. Clients obtain one significant launch per quarter.
The required success charge throughout 2.2 million automobiles is 99.9% or higher. Bell is exact about what which means: 0.1% of two.2 million continues to be greater than two thousand vehicles.
“I might by no means dare to press the button on 2.2 million vehicles and simply go,” he says. Bell doesn’t watch the 99.9%. He watches the 0.1%. And even inside that fraction, there’s a exhausting distinction. “A failed replace that may merely be retried: superb. An replace which means you may now not begin the automobile: catastrophic.”
When monitoring knowledge exhibits failures clustering round a selected market or car variant, the roll-out stops for that group whereas the remaining proceed. The issue is fastened earlier than that section receives the replace. Each deployment, Bell says, is a precision operation. And precision, at this scale, requires individuals who perceive precisely what they’re constructing.
Automotive coding abilities exhausting to come back by
The expertise downside is structural, and Bell doesn’t soften it. “Automotive software program engineering that we do now’s principally a five-year-old self-discipline at scale. All people’s seeking to rent leaders in software-defined automobiles. Nicely, there aren’t any.”
That isn’t a grievance in regards to the labour market. It’s a description of a self-discipline that hardly exists but. Writing safety-critical code for a two-and-a-half-tonne car below ISO 26262 purposeful security requirements just isn’t the identical as constructing client functions.
The context is totally different, the constraints are totally different, and the implications of getting it incorrect are totally different. “It’s not identical to hiring software program individuals right into a automobile firm. It’s not about figuring out the best way to code. It’s about coding within the automotive context.”
Volvo constructed that context itself. The toolchain, the testing infrastructure, the repository construction, the method for configuring one codebase throughout eight silicon generations and 4 factories: none of it got here off the shelf. “There was completely nothing after we began,” Bell says. “There’s no tier one that you may name and purchase an answer. We needed to create all of it ourselves.”
This explains the extent of inside funding that the majority organisations would wrestle to justify. Volvo’s software program take a look at centre in Gothenburg covers 25,000 sq. metres and price round €26m. And even then, Bell’s most vital lesson from the previous 5 years is solely this: “Begin earlier. Don’t underestimate how large this job is.”
HuginCore is the centrepiece of that funding, however it doesn’t make Volvo unbiased. The platform runs on Nvidia’s Drive AGX Orin system-on-chip and Qualcomm’s Snapdragon Cockpit Platform, and integrates Google for AI.
Bell acknowledges the stress with out pretending it away. “It’s by no means snug being dependent.” Lengthy-term agreements and contingency planning are the reply, not vertical integration all the way in which down. “Know-how evolves, and we are going to evolve with the expertise.”
The more durable query is the place to attract the road between proprietary and shared. Bell believes the decrease layers of the stack, cyber safety, replace administration and base infrastructure, will finally turn out to be commodities.
“Do it is advisable to do every thing your self, or do you purchase bigger chunks of it and share that funding throughout a number of automobile makers? I feel that might make sense,” he says. That query has not been settled in automotive, and it isn’t settled in enterprise IT both.
Deal with venture tempo and preparation
For expertise leaders outdoors the automotive sector, Bell’s core lesson just isn’t technical. It’s about tempo and preparation. “Breathe deep earlier than you go in. Don’t attempt to do an excessive amount of too rapidly. Don’t put your venture on an aggressive timeline for job one. Do the basic proof of idea earlier than you decide to the product plan.”
He additionally factors to organisational dimension as an element that’s not often mentioned truthfully. Volvo is, in Bell’s description, in all probability the smallest unbiased automobile growth operation within the Western world. “We’re sufficiently big to grasp the scaling you want. However we’re sufficiently small to not be institutional, to not get caught in committees.”
That stability is troublesome to duplicate, however the underlying precept applies in any organisation the place software program transformation has to occur at velocity, below exhausting constraints, with penalties that reach past the display.
In monetary companies, healthcare and significant infrastructure, the self-discipline is identical. “We will’t have a look at anyone and say, ‘Hey, we must always try this’,” Bell says. “We have to comply with our personal path and stroll the trail that no person walked on earlier than.”
The diamond just isn’t completed, however it’s being polished.
