Elon Musk emphasized the opportunity for distributed AI inference using the fleet of Tesla cars and the future fleet of Teslabots.
I tried to characterize the details of distributed AI Inference with the future fleet of Tesla cars and Teslabots.
I have a forecast of the numbers of car and bots in each year and the specifications of a 10X over Hardware version 4 for Tesla AI5 chip.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Do the owners get paid for the use of hardware they bought, and power they’re buying?
Good overview.
Feedback:
1. I don’t think everyone will sign up for making money by outsourcing their car chip for revenue. Maybe only 30%.
2. With the power useage, it might be only cars that are charging will be used. The power useage of AI5 is estimated at a few hundred watts. That is not negligible. It means that the car would lose ±15 – 30 miles of range per day. After a few days, that’s too much. I think estimate only ±10% of cars.
3. I think The Tesla Bots are over estimated.
Combining this, at a fleet of 10M cars, only 300.000 can be used for inference at the same time.
300.000 * 2500 dollar = 750M$ revenu.
It is not insignificant, because it would be almost all profit. Comparing to profit Tesla now makes, it would improve profit by say ±5%.
What Tesla sneakily could do is add a 1kWh backup battery for AI alone, and tell customers car battery will never be used. It would be only $100 costs for Tesla per car extra, and only add maybe 5 kg. That way, range is never impacted / there is no range loss.
This could increase the % of customers going for it, and increase the % of cars that can be online.
A 1 kWh might be enough for maybe 5 hours of inference. Cars that are daily charged could see an uptime of 12 hours over night, and 5 hours during day.
Some more calculations:
1. Assume 30% of US fleet is daily charged (Europe people usually don’t have their own charger). Assume 80% of the customers are going for it.
2. Assume fleet of 10M cars in the US
3. Assume 17 hours of online time per car per day.
2500*10M*0,3*0,8*17/24 = 4,25 B$ of revenue per year.
If that is almost full profit, that would be about 25% extra profit vs Tesla 2023.
I doubt the cars have enough memory to run even the AI models they were planned to run when designed. Running additional software that sucks up all memory and bandwidth will be bad. Running FSD AND other software simultaneously is even more unrealistic.
They will have to do this when the cars are sitting still charging for long periods and also have access to high bandwidth.
Good concept but the current hardware isn’t ideal.