Tesla humanoid robot

[Buckminster]

$10K hdw sounds fine, but that's a true COGS. Any software developement will come from R&D which is spread across mulitple product lines, including FSD. The accounting will be different, I think. Paging @st_lopes


Yeah, 3 KWh is a good est for bty pack size, raw materials more like 1/30 of a model 3 (125 lbs x 20 = 3750 lbs)


Charging 3 x 1 hr per day, uptime ~20 hrs/day. As long as the bot is available for work while the factory is running, Teslabot fits in fine at the factory (not a bottleneck).


Interesting idea about the work-for-hire, but Tesla will have to sell at least some to cover production costs, depending on their margins. $3-$5 is WAY low-ball: Tesla said the bot is to perform

1. dangerous tasks (read: Fukushima cleanup, Fire/Rescue/Emergency responce, Bomb disposal), and
2. for jobs that people don't want to do (Coal miner, TBM Operator on Mars).

I'd estimate $125/hr for the former, and $25/hr for the latter types of work. The main cost difference will be MTBF in different tasks (likely firefighter 'bots earns ↑$ while Fuku gets ↑↑$)


Agreed. Informative. Indeed.

Sky's NOT the limit anymore...

Cheers!

Sandy Munro estimated the cost of labor for assembling a $50K Model 3 at about $10K.

With Teslabot, that drops to about $1K in the medium term (3-4 yrs).

That takes COGS to about $24K and gross margin to about 52%

Excited yet?

 

[Buckminster]

Quick note I didn't see called out here:

There was this blurb about Karpathy posting some Tesla AI roles in his team. The bit that caught my attention was that Tesla Bot was described as an AI platform. This sent all sorts of bells ringing in my head. This may be very familiar for the "Product" people, but for those that are not familiar, a Tesla Bot platform implies an iPhone like capability where other developers can code the bot to do specific things, and sell those capabilities via an app store. And tesla will take a cut of the revenue.

Of course Tesla will have its own apps for the bot, and open source some reference implementations. And there will also be a digital model of the bot and a hundreds of thousands of stochastically generated digital environments in which other developers can train the bot to do useful stuff. Like entertaining your kid with a game or rock paper scissors, or pick up toys, or fold the laundry, etc. The complexity can ramp up over time. And the guts can be modular and upgraded.

This is one of those things that will have massive network effects where a critical mass of consumers and developers come together, and when that happens it becomes almost impossible for a newcomer to unseat the incumbent. Again like the appstore. At some point an Android equivalent may come along, but I am not sure if anyone is even thinking about it.