In 2019, Tesla announced with great confidence that it had developed the most powerful semiconductors for self-driving vehicles and claimed them to be superior to the ones by Nvidia which it had used for a while for its AutoPilot mode. But now comma.AI CEO George Hotz believes that Tesla should actually sell its chips to its rivals and create a more viable competitor to Nvidia and make some money while doing so. Comma.ai develops self-driving software just like Android is to smartphones. Hotz famously once was even offered a contract to develop a self-driving system for Tesla. His start-up makes software that runs on Nvidia's chips but theoretically, his software can also run on Tesla's chips considering both chips are based on the same ARM architecture.
Tesla had claimed at launch that its self-driving chip was 21 times faster at a frame per second processing while only increasing power consumption slightly. Tesla had a rockstar team in place with people like Jim Keller who has had a hand in developing Apple's A4 chip for the iPhone 4, AMDs recent CPUs and more recently was a senior vice president at Intel.
Hotz believes that Nvidia's monopoly can be broken if Tesla were to supply its chips to third parties. Tesla, of course, is following an Apple-like vertically integrated strategy in which it believes its chips are its secret sauce. Notably, using the Dojo computer and without then use of Lidars, Tesla has been able to deploy full self-driving capabilities.
“You basically have two options right now to train. Your options are Nvidia or Google — and Google is not even an option since their TPUs are only accessible through Google Cloud. Google has absolutely horrendous terms of service restrictions. They may have changed it, but it used to explicitly say that you are allowed to use Google Cloud ML for training autonomous vehicles or doing anything that competes with Google without their prior permission,” said Hotz outlining the limitations in the market.
Hotz's reference to Google is about its Tensor Flow Processors which run in the Google Cloud.
"They should sell that chip (Dojo) and even their accelerator (HW 3.0 computer) that is in all the cars. Sell it. Why not? If you sell the chip, here's what you get: you make money off the chip. It doesn't take anything away from your chip and the world is going to build an ecosystem of tooling for you,” he added. Before concluding he also said, “Somebody has got to take Nvidia down.”
This also comes with the context of Nvidia's recent acquisition of ARM holdings which makes the underlying processor architecture that is prevalent in smartphone, tablet, wearable, 5G and self driving semi-conductors.