We examine Tesla, the king of disruptive innovation.
We recap our core theory, examine Tesla’s lead in battery technology and plans to scan production. We also examine its hardware strategy and data advantages as it increases the autonomy of its cars.
Why Tesla? They are unarguably the disruptor in the transition to EV’s.
They create the most efficient EV’s getting more range out of a KW of electricity than any competitor so far. Because most cars will be owned by fleets, that efficiency and therefore lower cost base will drive fleets to buy teslas as they will be more profitable to run.
They have the most advanced plans to scale battery production. They already have the gigafactory manufacturing around half of the world’s total output of lithium ion batteries. They also have a roadmap to scale to 1-2 terrawatts of production.
But what about the AI part?
First you need the sensors to collect all the data that the AI needs to drive the car. Teslas are all built with 8 cameras, ultrasonics and forward- facing radar. Tesla has chosen not to use LIDAR unlike all the other companies trying to solve autonomy. It’s clear that 8 cameras plus radar is sufficient as humans do it with two cameras and no radar (which can see through fog, snow, etc). Also, the car can look in 8 directions while simultaneously checking the map and never gets tired or dozes off, loses attention while talking, texting or whatever. Elon is not anti-lidar, he just doesn’t think it makes sense in cars. Given his track record it would be a brave (or perhaps foolish) person who says he’s wrong.
Second you need the compute power to process the data for all those sensors and make inferences in real time. Tesla started designing its own chip over 3 years ago. All new cars have hardware 3 which has 144 teraflops of processing power plus built in redundancy
In the year 1997 1 teraflop cost $50m. Now Tesla have 144 teraflops for a few hundred dollars and hardware 4 will no doubt improve on that by at least 3x – hardware 3 is 21 times faster than the Nvidia-based hardware 2 it replaced.
Machine learning requires huge amounts of data to learn how to recognise the millions of different.
Because the sensors are in Teslas fleet already, they are collecting billions of miles of data already and that collection is growing exponentially. The nearest is Waymo with 20 million miles, but most of those are within a designated area of Phoenix so they don’t contain the variety that Tesla’s data does.
Tesla drivers are training the AI. Every time a human overrides autopilot, the data from that incident is used to learn how to deal with it better the next time it happens.
EV driven by AI is a very different problem to those that traditional auto companies have solved over the last 100 years. The AI is a data plus software engineering problem.
Tesla has access to the best engineers available, very few of whom want to work at BMW or Ford
EV driven by AI is a platform. Only two or three such platforms will see mass adoptions. It’s more like iOS and Android or Windows, Mac and linux. Given their lead it seems almost inevitable that tesla will be one of those 2 or 3. Remember, disruptors come from the outside. The other main platform may well be something like android – hardware manufactured by VW or Ford and software from Google.