Elon Musk wants to turn Tesla fleet into AWS for AI – would it work?

During last night’s earnings call with investors, Elon Musk floated an all-time idea: what if AWS, but for Tesla?

Musk, who loves to discuss results, compared the unused computing power of millions of idle Tesla vehicles to Amazon’s cloud services business. If they’re just sitting there, he thought, why not put them to good use to run AI models? (Also, have you ever actually looked at your hands? No, I mean it really looked like?)

“There is a potential… when the car is not moving to actually perform distributed inference,” Musk said. “If you imagine the future, maybe where there is a fleet of 100 million Teslas and on average they have maybe a kilowatt of inference computing. There are 100 gigawatts of inference computing, distributed across the world.”

So, to summarize, you buy a Tesla. It’s your property. But Musk wants to freely use the unused computing power in his vehicle for… something? Possibly related to AI? Hopefully not blockchain. (By the way, Tesla is now an AI company. Musk himself said as much during the call.)

Would Tesla pay for this? It is not clear. After all, this is Musk at his most hypothetical. Still, I wouldn’t doubt that he would just try to take the computing power out of his customers’ vehicles without consent or compensation. GM was giving your driving data to insurance companies without your consent! Baby, it’s a free-for-all.

But before we can even treat this as a serious idea, we need to find out if it’s even possible. I reached out my hand to Sam Anthonyformer chief technology officer at Perceptive Automata, a now-defunct company that built modules for self-driving cars to allow them to perform “theory of mind” tasks.

Anthony said that, as a concept, it is “perfectly possible” to split large computing tasks across multiple small nodes. We’ve already seen this happen with Bitcoin mining or Foldable@home, a distributed computing project to develop new therapeutics. But just because something is possible doesn’t necessarily mean it’s a good idea.

Anthony said there are two main problems that make cars – and electric cars, in particular – imperfect nodes for a distributed computing project. First, you need to rely on your car’s battery or, if it’s plugged in, the charging station’s power source for power. And this energy is often not free, with homeowners paying retail rates for electricity. Second, connectivity and speed are a “big issue” in distributed computing, Anthony said.

“Inference, in particular, is part of the [machine learning] workflow where your speed is essential,” he added. “You’re not doing a lot of inference offline overnight, you’re answering questions as they’re asked – that’s the big inference problem that AI companies are facing right now – which makes connectivity issues and Availability of cars (which, you know, move around) is even more of an issue.”

In Musk’s opinion, the distributed network would only work when cars were parked or immobile. Still, Anthony argues that no one would voluntarily create a distributed computer architecture from millions of automobile ECUs (electronic control units) unless they were somehow forced to do so.

“It’s someone with a very strange-looking hammer imagining the existence of deeply implausible nails,” he said.

In fact, computer scientists have been trying for a long time to create fast computers from many small, idle nodes. One of the first examples was SETI@casa, in which Berkeley researchers thought they could find extraterrestrial life by accessing a volunteer network of distributed computers to analyze radio data. So why not a Tesla@home?

For one, the more geographically distributed the nodes, the harder it will be to get them to work in conjunction with each other, said Phil Koopman, professor of electrical and computer engineering at Carnegie Mellon University. who co-authored a book about supercomputers.

Like Anthony, Koopman recognized that the project could work as long as the vehicles were plugged in during computation to avoid battery drain. Good Wi-Fi was also a necessary component, so the Tesla would likely need to be parked at home overnight for the distributed network to work properly. But even then, you will likely encounter obstacles as you develop the project to make it useful for AI computing.

“Scaling to this size is always challenging and rarely successful enough to make it worth doing this rather than building a data center,” Koopman said. “The devil is in the details, so I would like to see some serious experimental confirmation that it is viable.”

Musk loves to pontificate about what will be possible in a future dominated by connected autonomous vehicles. Things like a 24/7 robotaxi service where your vehicle earns passive income while you sleep sound incredible in theory. But when the rubber meets the road, Musk’s big ideas tend to fizzle out.

“For now it’s an interesting idea,” Koopman said, “but we need to keep in mind that most interesting ideas like this are not intended to be practical.”