Southpasfan
Member
I'm curious. So take a look at the intro to Anthony Bourdain's "A Cook's Tour." A bot slowing folding a shirt to me looks to be many more years away from actually working in a functioning kitchen than the current FSD is from driving without a human.The Bot Will Drive Future Valuation of Tesla
I've been pondering over this for a few months. We all likely came to Tesla years ago to route for the acceleration of a path to a sustainble energy future via electrified vehicles and energy products. That path is well on it's way now. We then looked excitedly at adjacent ventures, mostly robotaxis, as an exciting path to grow the company's value while aligned with greenfiying the economy.
But the path toward a software stack that is at a high enough level of fidelity to "turn on" for robotaxis nationwide may still years away. While FSD progress has been good and should improve with iterations of V12, we can't ignore the rigorous statistical nature that is required to be true robotaxi. While the FSD Beta Tracker estimates of about 100 miles / critical disengagement may be biased, we know it's not near the 10,000 miles / critical disengagement we need to get to to allow the car to work as a robotaxi. My best estimate is that the software needs to reduce interventions / disengagements by a factor of 10x at minimum, if not 100x to reach that utility.
I do believe that is achievable, but on what timeframe is less clear. I do know that I will rely more on quantitative metrics too the progress than simply listening to Youtubers alone. No bias in entries can mask a 10x improvement in rates. I am excited to see this happen over the next year or two. This can add a trillion dollars in valuation.
But what I've realized is that Tesla can deliver a subset of the eventual utility of the Bot in a much shorter time frame than we may be expecting. As I did my graduate work in modeling humanoid robotics, I am explicity attentive to how well it will learn a robust dynamic balancing policy to handle diverse environments. This is essential for allow the Bot to be used at a wide volume and scale of tasks.
But, does that even need to be achieved for a company to use the bot in a profitable manner?
Let me attempt to "skate where the puck is going".
Tesla has already illustrated a highly competent hardware package, even in the video shared yesterday.
We know even outside of Tesla, that there is an acceleration of progress on deep neural nets for learning and executing a range of manipulation tasks. These bots will be able to watch a human perform a task a series of times, then be able to generally replicate it.
When I look at the hardware and software stack, and integrate it with what Tesla's AI team will be able to execute in a year or two from now, I see the Bot being able to perform a decent variety of tasks, even if limited in locomotive capabilities.
Imagine, for instance, if the Bot was only trained to perform a set of tasks in a fast food restaurant. Identifying the types of pieces of chicken, combining the various sides as per what was ordered, and putting it into the bag and giving to the customer. The perception and dexterity capabilities are already there. The bot may not be able to perform all tasks in the restaurant, but I bet it could handle 1 out of 4 employees.
Employees are expensive. In California, minimum wage is $15 / hour, so for 12 hours a day, you will spend almost $70k per year for that work.
A company would easily pay $35k / year for a Bot to replace that human work.
There are 200,000 fast food restaurants in the U.S. alone. If they are purchased for $50k with 50% gross margins, that's 5 billion dollars in gross profit. We haven't even talked about any other works in the restaurant, let alone other jobs.
The potential profit of the Bot is simply orders of magnitude higher than robotaxis, and the threshold performance is lower for many jobs. You can have a Bot that ruins 5% of the food and still be massively profitable. A 5% hiccup in a car is unacceptable.
Given all this, it seems to be the Bot might be able to start generating revenues and profits before any robotaxi services. In fact I predict it. I believe Tesla Bot will generate a billion dollars in profits before robotaxis, and never look back.
And when this happens - maybe 2-3 years from now - the market will being valuing the future growth of that market and the numbers could be staggering - adding multiple trillions to the market cap.
Working in a kitchen not only requires the same visual to computer to action processing that the FSD computer does, but on a much more delicate level. Dicing an onion at the level of a graduate of a culinary academy is a matter of fractions of an inch. A car does not need to be controlled to that degree of specificity.
Not only that, but unlike the FSD computer, the bot will have to have auditory processing which may be easier than the visual processing but nevertheless has to be added to the mix.
Yet you think the bot is closer? There are already industrial robots which can, when fixed, handle millimeter like physical processes, I would imagine on the Tesla line the spot welding does this already. But a bot has to first orient itself in space via sensors before it can do the same thing. Right now after years of work the FSD suite is having trouble orienting itself within feet and yards.
Isn't the spatial orientation of a commercial kitchen some orders of magnitude more difficult that merely driving a car and not hitting objects?
I get the math on the market. I am not sure I see the ease of development from here.