Tesla is just one of many automakers and third-party tech companies looking to solve self-driving. While some automakers have chosen to offload that responsibility to companies like Mobileye, others like Tesla have decided to tackle the problem head-on.
Now, after years of throwing money at its own hardware and training stacks, Tesla has amassed a gigantic multi-billion dollar investment. And in doing so, it figured out something that nobody should be surprised about: building a car that can drive itself is very, very expensive.
Get Fully Charged
Tesla's Race For Autonomy
Tesla has been aiming for autonomous driving capabilities for years. Since 2016, Musk has been promising a fully autonomous cross-country drive in one of its vehicles, though that has yet to happen. Now, as other automakers and autonomous ride-hailing services advance, Tesla will need to sharply increase its hold on the market or face the threat of its self-driving tech being bested by competitors.
In a post on his social media platform, X, Tesla CEO Elon Musk revealed that Tesla is on track to spend more than $10 billion on computing, storage, and networking solutions used to train the model for Tesla's Full Self-Driving software.
Now, this number is cumulative, meaning that it's the total amount that Tesla has spent over an unspecified period of time. It's worth noting that the $10 billion figure likely includes legacy systems powered by Nvidia GPUs as well as Dojo, the automaker's supercomputer running on custom Tesla silicon. Last year, Tesla said it expected to spend an estimated $1 billion on Dojo.
Musk has long been a proponent of investing in computing power to advance the model used by its fleet of semi-autonomous cars globally. By investing in hardware capable of processing video clips and telematics data sent by vehicles at a more efficient pace, Tesla's model can quickly "learn" driving decisions much more quickly.
"The fundamental rate limiter on the progress of full self-driving is training," said Musk during the company's second earnings call of 2023. "If we had more training compute, we could get it done faster."
Shortly after it began enabling a free month of its Full Self-Driving software for all hardware-equipped vehicles in the U.S., Tesla announced that its fleet has surpassed 1 billion miles of driving on FSD. This works out to an investment of approximately $10 per mile driven.
Musk has also said that Tesla plans to reveal its long-awaited robotaxi on August 8th. If Tesla plans to usher in any form of Level 5 autonomy like what might be needed for the robotaxi, it needs to, at minimum, seriously scale its training data. And $10 billion buys a lot of power to do exactly that.
