is Tesla an AI Company?

Since ChatGPT was released to the public in late 2022, the understanding of AI’s potential is spreading. The impact on the economy is expected to be magnitudes of trillions of dollars over the foreseeable future. Wall Street is on board now and the market is starting to explode regarding industries using AI. AI vendor Nvidia’s stock is meteoric and Tesla’s (a self-sustained real-world AI-based company) stock is gaining steadily.  download article

An AI-based company is different than an AI vendor. Tesla develops:

  • cars that use artificial intelligence to self drive
  • supercomputers, named Dojo, that learn to see and understand the world around them
  • humanoid robots that use artificial intelligence to navigate and do things
  • battery power stations that support the power grid using artificial intelligence 
  • in-house AI systems that control activities in the factories

We say “real world” AI to mean the ability to interpret video into perception and likely response to situations. Humans may misinterpret what they see or not know how to react to surprises on the road. A self-driving car is not allowed the privilege of mistakes. Driving at 70 mph with a myriad of moving objects, road conditions, pedestrians, and traffic signals, requires an onboard supercomputer trained by a supercomputer at the factory.

The key to Tesla’s success is operational efficiency and continual innovation. Innovations in manufacturing technology have springboarded their success. That being said, it is the real-world AI that positions the carmaker to be the world leader in cars, humanoid robots, and AI-based systems.

One advantage is they are mostly self-reliant. Software and hardware is in-house and proprietary. Operational systems are all homegrown. The AI hardware was developed at Tesla which is used for self-driving and the Tesla humanoid robot. 

Another improvement is the factory is managed by data instead of people. Instead of managers, there are engineers that continually improve the line and products. Mob teams use the system as a sidekick. The factory is self-driving. The innovation occurs incredibly fast because changes happen in nanoseconds instead of months.

Mobs are cross-functional groups assigned to solve issues or implement improvements. Improvements are fed back into the system and AI algorithms place a value on the efficiency gained. The value is then assigned to the product. Idea failure is expected such that there is no fear of being creative. So, the improvements come fast and furious. The velocity of innovation is astounding as a result.

There are data boards that direct the mobs as to what is happening and where to be. Dedicated mob teams working 4 am to 7 pm run by computers are impossible to beat. Communication is by proprietary chatbots run by AI.

a picture of Tesla self-driving

Tesla’s full self-driving is the number one real-world AI tech in production. This means Tesla is a leader in real-world Artificial Intelligence. They have pursued FSD for years. Further, their manufacturing and administrative systems are developed in-house and AI-based. Further again, the Tesla humanoid robot uses the same visual AI tech as the car and the AI solves the robot’s functions.

The significance of the full-self-driving, also known as robo-taxi, is the revenue potential. A robo-taxi will take in fares without paying a driver. That is an enticing proposition that is likely to disrupt the Uber taxi service.

However, let’s be realistic about robotic cars and manage expectations. Self-driving cars have to deal with emotional human drivers and passengers. The solution to the driving algorithms is more complicated than just following the speed limit and avoiding accidents. Humans won’t tolerate this complicity when we are in a hurry- say we are late for an appointment. FSD cars have to deal with road rage drivers and impatient passengers. The AI engineers really have their work cut out for them. The good news is the millions of miles of driving data that Tesla cars have logged. This vast amount of data used by driving algorithms put Tesla years ahead of the competition.

a picture of the tesla bot working 

The importance of the humanoid robot is that it is capable of doing mundane production line tasks. It is estimated that the bot will do about half the labor in the Mexican factory once it is up and running.

The reason the Tesla bot is newsworthy is the cost savings in labor at the Tesla factories. The value proposition is compelling.

 The bot’s minimalist design allows Tesla to pump out robots by the thousands if need be. The bot uses the same computer and batteries as the car. That’s one win for Tesla. The actuators and all other parts are easily made at Tesla. The bot already can walk, carry objects, grasp tools, and perform tasks. 

What is amazing is within a year, the engineers solved the two most difficult problems- balance and hands. Then after another six months, taught the bot to work. The Tesla bot is not science fiction but is in fact a reality.

The humanoid robot is a game changer. All the parts are easily made at the Tesla plant. It uses homegrown actuators and modified tech from the car business. The bot has visual acuity borrowed from the self-driving tech. The hands are an innovative simple design. The digits are all the same but mounted on an arc. It can carry 6 kilos in each hand from the bin to the production line.

From an investor perspective, the potential savings from the bot and revenue from full self-driving is exciting news.

The philosophy of Tesla is raw materials in – products out. You may call it Vertical Integration where most things are made in-house. Since Tesla is mostly self-sufficient and AI-based, they have a unique position for efficiencies of scale. It costs less for Tesla to make things. Therefore, their profit margin is greater than most. This extra cash allows them to add factories and grow business at unheard-of rates. Of course, it takes time to build and furnish factories, but Tesla has a stellar record at doing the unthinkable.

In more exciting news, Tesla is planning to roll out a car insurance business. Yet another revenue stream to strengthen the car maker’s position. Although the premiums are not based on AI algorithms per se, Tesla uses AI driving data to determine the driver’s safety score.

To catch up, you can watch podcasts on YouTube at “Brighter with Herbert” where he talks with the experts on what’s happening with Tesla.

As of June 2023 Tesla can:

  1. Produce electric cars at a phenomenal rate and cost efficiency
  2. Make Full Self Driving cars using Artificial Intelligence
  3. Make functional Humanoid Robots 
  4. Install AI-based power stations
  5. Sell AI-based Car Insurance

FYI

Tesla’s Dojo supercomputer is the world’s 5th largest supercomputer.

Tesla’s supercomputer, Dojo’s specifications:

“Tesla’s Dojo supercomputer is a custom-built platform designed from scratch for AI machine learning, with special emphasis on video training using video data collected from all of its vehicles .

The Dojo V1 system has 53,100 D1 cores, is rated at 1 exaflops at BF16 and CFP8 formats, and has 1.3 TB of SRAM memory on the tiles and 13 TB high-bandwidth DRAM.

The power and cooling are integrated directly on the Training Tile, which is capable of 10 TB/s on-tile bisection bandwidth and 36 TB/s .

The Dojo D1 chip provides stunning bandwidth and compute performance, with 1.25MB of SRAM that it can use as working memory with five ports, but it has no cache or virtual memory.

It uses DMA to get the data in and out of the chip. The Dojo supercomputer is expected to boost Tesla’s capacity to train neural networks using video data, which is important to the company’s computer vision technology that’s working behind its self-driving capabilities.”

Leave a Comment

Your email address will not be published.