How are software-defined vehicles changing the automotive industry, and how is Arm involved in this shift?
The automotive industry has been undergoing a major evolution for the past five years, with new software applications now being deployed in cars. The expectation is that everything has to work immediately - unlike with a smartphone app, where if something does not work, you can delete it. So, the entire process of software development in the cloud and its deployment in vehicles is changing. To push new experiences or advanced driver assistance features into the car, a complete software-defined infrastructure must be built—both in the cloud and in the vehicle.
Another major change is the growing demand for compute performance in vehicles as new features are added. This is why, for the first time, Arm introduced a new grade of data-center class processing to automotive with Arm Neoverse. NVIDIA DRIVE AGX Thor compute system is the first product to leverage this new class of compute. Neoverse allows us to step up in performance from Cortex-A class CPUs to continue to meet the industry's compute needs.
Why do we need more compute in cars, and what kind of in-vehicle generative AI applications will that enable?
If you want to run Gen AI, like new driver or passenger monitoring applications, you need these to run in the car. For instance, with driver monitoring, you do not want latency from sending data back to the cloud and waiting for a response. If a driver is drowsy on the highway, real-time action must happen in the car, which means the compute and software have to be embedded within the vehicle itself.
With SLMs and AI agents coming in, we are seeing smaller models tailored to specific applications. One of our partners, for instance, is working on replacing the user manual in cars with Gen AI. If your car breaks down, instead of reading a manual or calling for help, you can talk to your car: “What does this warning light mean?” and it will explain the issue and suggest next steps.
Beside providing enhanced compute, how is Arm helping carmakers to integrate and capitalize on new software capabilities?
Powerful compute infrastructure needs to be supported by the right software tooling, compilers, and capabilities. Kleidi – our open-source AI library that optimizes machine learning and computer vision – supports developers to enable these applications efficiently on the CPU in the car.
You travel to tech and automotive trade shows all over the world. How do you evaluate the SDV market in Europe compared with other regions?
Let’s talk about China compared to the rest of the world, where the volume of the car industry lies, and Europe - particularly Germany - North America - particularly Detroit and the Bay Area - and Japan. These are the traditional car makers, and they rely heavily on the legacy infrastructure they have built over the years—tier ones, software ecosystems, silicon players. When it comes to software-defined vehicles, the key is unlocking new experiences and features in the car dynamically. But if a car has 100 ECUs and 80 of them are not software-managed, you cannot push new features or experiences easily. So, OEMs in Europe, North America, and Japan are now trying to simplify the complex software architecture they have built over decades.
In contrast, new OEMs entering the market are moving fast because they do not have as much of that legacy. Tesla is a good example – they did not have legacy supply chains or compute architecture, so they could start fresh with simplicity in mind, knowing they needed to enable a new software infrastructure. Without legacy, it is easier to innovate – but legacy also provides the experience to innovate, so it is a chicken-and-egg problem. That is the shift happening now: OEMs in Europe, Japan, and North America are realizing they need to simplify architectures and build effective cloud-to-car infrastructure. Meanwhile, development speed for new OEMs in the market – many of which are based in China - is happening at a fast pace, and it is pushing the rest of the industry to rethink how to simplify development cycles, bring software forward in the process, and enable software before hardware is available.
How does that play into Arm’s strategy?
We have a 35-year legacy in the automotive space, mainly supplying core compute complexes with the highest levels of safety, security, power efficiency, and real-time capability. But over the last five years, we have been looking beyond supplying IPs. In 2019, during a visit to Japanese OEMs, we received feedback: while they acknowledged Arm’s strong ecosystem, they wanted consistency in software development across all Arm-based hardware. Samsung, Qualcomm, Renesas, NVIDIA, for example, all have subtle differences in chip behaviour and debugging, which complicates software standardisation.
That feedback led to the creation of the SOAFEE Consortium with our partners. Our legacy is IP, but now we are moving up the stack, enabling real standardisation in the non-differentiating layers of software—leaving room for differentiation where it matters. We are also collaborating with cloud service providers like AWS to support the software-defined vehicle vision, including virtual platforms in the cloud to simplify software development. It is about an end-to-end solution; in automotive, you cannot just offer hardware or a chip without tackling software complexity.
