Thread by @The_AI_Investor

A typical car company:
- may make engines
- most parts, software made by suppliers
- dealerships do sales & services
-> mainly assembling

$TSLA: the above, and
- electric powertrains
- battery packs
- super charge network (= gas station)
- Tesla OS software
- AI chip
- FSD

$TSLA moats:
- best manufacturing
- best electric motors
- best batter pack density
- largest supercharger network
- best data for FSD
- vertical integration -> faster rate of innovation
- best CEO that can drive product innovation
vested.co.in/blog/tesla-str…

$TSLA's manufacturing
- Tesla's factory: machines that make machines
- factory is the competitive strength of Tesla long-term
- giant casting machines -> make cars in the same way that toy cars are made

➡️ exponentially growth in production rates
analyticssteps.com/blogs/manufact…

These are the reasons why Tesla dominated the list of most American-made cars (2022).

Tesla benefits from being highly vertically integrated while the rest of the industry has focused on vehicle bodies, assembly, and engines while relying on suppliers for much of the rest.

This year, Tesla is focusing on scaling up EV to capture market share while demand is skyrocketed. With more cars on roads, $TSLA has better chance of achieving FSD. FSD then leads to robot taxi and AI robots.

➡️massive value added to Tesla car owners (and $TSLA share holders)

To solve FSD, Tesla will need to solve real world AI:
✅silicon neural nets (brains)
✅cameras (visions)

A robots on 4 wheels problem (FSD) can be generalized into robots on legs.

Tesla humanoid robots can be bigger than car business, according to @elonmusk.

$TSLA is transforming into Robotics & AI company, while competitors are still having a hard time scaling up production.

✅Tesla - converging of AI, Robotics and Energy Storage

$TSLA FSD latest updates:
- we rearchitected the neural net for 1000 times
- radar & ultrasonics were a mistake
- 2 things are taking most amount of Elon's brain space: self-driving and star ships to orbit
- Tesla will probably have at least 5 year lead over competitors

More technical resources from Tesla AI Day 2021:
towardsdatascience.com/tesla-ai-day-2…

The above moats translated into super high margin, $TSLA is on its own league comparing to other car makers.

✅ $TSLA >= $AAPL for integrated hardware & software products that people love.
✅ $TSLA >= $TSM in manufacturing efficiency and high margin.
✅ $TSLA >= $GOOG in AI.

$TSLA's gigafactories, why it's so hard for competitors to catch up:
- the factory is the product
- most advanced car factory the earth has ever seen. Alien technology (@elonmusk on Giga Texas)
- the factory is like a chip, raw materials in one side, cars out the other side

$TSLA's manufacturing expertise in batteries as competitive edge:
- eliminate steps
- streamline processes
- slash costs
- battery factories next to car plants and chemical plants
- Tesla’s new 4680 battery cells are far cheaper and can store far more power per unit of volume.

$TSLA high margin:
- lower batter cost
- direct sales
- no marketing expense
- simpler design with fewer options
- scale automation in manufacturing
- software, FSD

"People who are really serious about software should make their own hardware." - Alan Kay

$TSLA has been very serious about software and vertical integration, that why it overcame the chip shortage while legacy car makers struggled.

The unified computing architecture made it possible for Tesla to rewrite its software to work with chips that were not in short supply.

$TSLA also took control of their FSD destiny by replacing $NVDA with its own AI chip (manufactured by #Samsung).

@elonmusk, 2019: Telsa designed the best chip in the world, best by a huge margin. Tesla claimed a 21x gain in perf over the previous Nvidia's chip & with 80% cost.

How it was possible for Tesla to design the best AI chip, while it had never designed a chip before?

That's because of Elon's clear vision on what he wanted, and the 'how' was solved by legendary chip designer Jim Keller.

en.wikichip.org/wiki/tesla_(ca…

$TSLA also built its own NN training supercomputer, Dojo, with its 2nd chip, the Dojo D1. The chip was designed by Tesla all the way from the architecture to the package.
✅best AI training performance
✅enable extremely complex neural net models
✅power and cost-efficient

$TSLA has built the complete integrated ecosystem for its FSD from battery, car design, advanced manufacturing at scale, software, chip, and AI training.

Take a step back to see where FSD is in Tesla's complete energy and transportation ecosystem.

$TSLA Dojo could become the best super computer for AI that ever developed.

Tesla will use Dojo to train its neural networks FSD, but as mentioned by @elonmusk, Tesla could open it up for other developers as well, and that could potentially become a new business for Tesla.

$TSLA's FSD focuses on solving computer vision problem, which is different from most other companies like Waymo that also use radar, lidar, etc.

@elonmusk famously said: Lidar is a fool’s errand.

@Mobileye creates 2 independent models, one from cameras, one from radar & lidar.

$TSLA FSD has another big, major advantage: real world data, and quality of data.

Telsa's cars are everywhere and increasing rapidly. This leads to the flywheel effect that will further distance Tesla from competitors, just like how $GOOG is dominating the search engine market.

Why Lidar is doomed:
❌expensive
❌useless in bad weather conditions
❌power hungry
❌ugly
❌can't differentiate objects
❌needs HD maps

With data, $TSLA can replace Lidar.

✅Competitors are wasting time and resources
➡️ $TSLA is now at 5+ years lead in AV.

Why it's possible for $TSLA to remove Lidar and radars?
- object detection, with depth, velocity, and acceleration by deep learning (supervised learning)
- NN learns to detect objects and their associated properties
- a combination of auto & manual (human cognition) labeling

$TSLA deep learning:

@lexfridman: “For Waymo, deep learning is the icing on the cake; for Tesla, deep learning is the cake.”

DL needs data.

All new Teslas are equipped with the FSD hardware. Tesla in fact filed patent for sourcing self-driving training data from its fleet.

$TSLA FSD vertical integration:
✅manufacturing cars and the hardware for FSD
✅video data from millions of cars
✅supercomputers to train deep learning models
✅AI chips installed inside cars
✅validates AI through shadow testing

Tesla owns the entire self-driving car stack.

$TSLA AI - while solving autonomous driving (AV), Tesla has become the leader in real world AI.

🔜Tesla is becoming an AI company.

✅Neural net
✅Autopilot
✅Dojo
✅Data
✅Simulation
✅Tesla Bot

Learn more about Tesla AI: