Chips ahoy: Why computing giants suddenly care about cars

Nvidia has three main automotive products: an in-car computer (Drive AGX), cloud computers that can be used to train AI models (DGX) and simulation computers that can work through driving models (Omniverse).

It has been a learning experience. “It’s tough, because the [automotive] industry is so bureaucratic,” says Kani. “There’s a political infrastructure with so many people involved. There are negotiations with governments and unions, the board of this and the board of that.

“That whole process of decision-making is new for a tech start-up. If we think something is right to do, we go and do it.

But automotive companies say ‘let’s kick off a proof of concept’, and they will work with you for years and you think nothing is going anywhere.”

Kani admits this was “a bit of a culture shock”, but there are signs that the automotive industry is slowly changing.

Until recently, car makers would release a finished product, with any revisions coming in a future mid-life update.

But now that cars are connected, they can be continually upgraded with new software. This has changed the way car makers think about computing – a change instigated by Tesla, believes Kani.

“In the past, if firms were building a cheap car, they would use the cheapest chip they could.

"For medium cars they would use a medium-price chip, and for high-end cars they would use an expensive chip.

Then Tesla put high-performance chips in all their cars and focused on software,” he explains. “With a software-defined vehicle, you put the highest-performance computer and richest software in the car and over the life of the car keep upgrading it.”

It’s the same approach, notes Kani, that Apple takes with its iPhones. And while established car firms were slow to react to Tesla, Kani says the rise of Chinese newcomers “building really good self-driving software” has changed things: “There’s clearly a threat there. So now they’re thinking: ‘How do we make decisions faster?’”

Many car makers talk about dramatically shortening their development cycles to try to match the Chinese – but Kani says “we haven’t seen that”.

He adds: “In China, they make a new car every two years. Most [other] manufacturers are on a four- or five-year cadence. If your competition is upgrading every two years and you’re doing it once every four to five, they’re running circles around you.”

Kani notes that the first European car makers to sign up to use Nvidia’s Drive AGX Orin platform (essentially a ‘brain’ for autonomous vehicles) – Mercedes-Benz and Volvo – took several years to get the tech onto the road, whereas Chinese clients “had cars in production using it nine months later”.

Those European cars on Orin such as the Volvo EX90 are just launching now, but Nvidia rolled out the more powerful Drive AGX Thor platform in 2022, and there are cars on the road in China using it already.

“So you started at the same time, but by the time you launch your car, you’re behind on day one,” says Kani.

“When we talk to Dell, HP and Apple, they’re asking when our next computer is available, and then the launch of their next product [an iPhone, for example] is aligned to it.

"What car firms need to start thinking is ‘when is the next platform available?’ and about how they can be ready [to start working with it] on that date.”

The car industry’s shift to software-defined vehicles hasn’t been easy. Look at the challenges faced by the Volkswagen Group or delays that hit recent Volvo model launches, for example.

Kani says that was partly due to car makers being slow to realise the type of software needed for the new generation of vehicles had changed, now putting a much greater focus on machine learning.

Asked for his opinion on how car makers are managing the shift, Kani says: “Certainly they’re getting more informed. Their teams are more and more knowledgeable.

"The challenge in the short term is the shift in how to develop a software stack. In the past, the likes of Volkswagen, BMW and Mercedes would take mobilised perception software, then do fusion software separately and then planning and control separately.

"That’s no longer how you develop software: you take that perception software you were outsourcing, and now it’s an end-to-end model that does perception, fusion and planning in one model.”

To strip away the tech terminology there: car firms used to develop individual bits of software to do specific tasks on a car, but the latest super-powerful chips developed by the likes of Nvidia enable far more advanced AI systems to be run, which is what’s needed for digitally advanced and truly autonomous cars.

“It’s doubling down on the know-how that they didn’t have,” continues Kani.

“The problem is that car firms did actually learn a lot, but the type of software they have knowledge of is no longer something you want to be using in the future, so they need to hire a whole bunch of new people who can do machine learning.

“It’s harder than everyone thought. But we will do our best to help them. That’s our job: ‘let us help you with these problems’.”

On that note, Kani describes Nvidia’s relationships with car makers as partnerships, with Nvidia tailoring solutions to meet car makers’ needs using its various chips and services.

While Nvidia’s in-car package is strong, Kani says that it’s in cloud computing and simulation systems where the firm is well ahead.

Indeed, its biggest automotive partner, Tesla, is now largely using Nvidia hardware for its AI development (CEO Elon Musk said he expected to spend $3bn-$4bn on Nvidia AI hardware last year).

The big business opportunity for Nvidia comes with self-driving cars: not just the level-two-plus cars on the road today (which can operate autonomously under supervision in certain situations) but the level-four (fully autonomous) ones the industry has dreamed about for years.