Self-driving cash flow

October 24th, 2024


It’s safe to say Tesla will not achieve level 5 self-driving before 2027 and the launch of the Cybercab. Elon implicitly acknowledged this in the “We, Robot” event by saying that people will own small fleets of Cybercabs and run them as small businesses. With just level 4, you need someone to 1. assume control during unforeseen circumstances (see Waymo “coning” or anytime a party tries to pack an Uber like a clown car), 2. assume liability, 3. maintain the fleet. For that reason, Tesla will need to provide remote drive capability in the Cybercab, likely over 5G (in cities and suburbs) and Starlink (in exurbs and rural). You assume some kind of VR headset solution, since any manual intervention will likely be at <20mph (Tesla stuck behind a fallen tree or broken stoplight).

Assuming they do that, Tesla’s approach gives them huge benefits in cash flow, which is vital in physical technologies. Tesla doesn’t need to build out a call center, buy the reinsurance, or deal with painting or vacuuming. Waymo’s solution might be technically superior, but if each car costs $300k, creating 15,000 Waymo’s (plus some spares) to replace NYC’s cabs will cost nearly $5b, plus building out the facilities and staff to support. But the manufacturing of the Waymo’s in any reasonable time frame is not easy either. If you’re looking to deploy 1m cars in next two years, you need to build out a factory, which has its own ramp time. All of this in a time when money has real cost (~5%). For consumers, cars don’t need a ROI: they’re a necessity. Your car just needs to pay for variable costs of FSD + charging (~$200/month) for it to be “worth it” to you.

As for Elon’s $25k estimate: the Cybercab will likely have a <200mi range and require a FSD subscription ($100/month) and likely a Starlink subscription too (another $100). Because batteries add real weight a car, cutting range by 30% lets you cut battery size by 50%. Of course, range doesn’t really matter if your car can go charge itself when you’re not using it. Lighter cars also reduce wear and tear on tires.

I think the exurb wedge is especially valuable. Cities have their solutions: public transit and Uber, but the further out you go, the worst the options because boredom is a real price to human idle time: Uber drivers need to make X money per day and just as importantly, you become frustrated if you’re online for 8 hours a day but only drive 2 hours, even if you net the same.

Finally, Tesla will inevitably shadow-run training when the car is not in use. It’s a question of power usage, but it’s relatively easy to for Tesla to grant cash or better yet, Supercharging credits to owners. Cybercabs could simultaneously charge and train, with power being “free” while training. Once again, this allows Tesla to sidestep some upfront fixed costs: Tesla owners pay for the GPU hardware and the public pays for the electrical infrastructure (in most countries, electricity is subsidized for the public).

Whether this all works or not depends on one big bet: that you can achieve level 4 without lidar. But given everything we’ve seen with LLM’s in the last two years, it feels like it should be possible. That being said, I don’t own any Tesla shares and am unclear how this translates into share price.