Deeper into the AI of Tesla and Nvidia

The global technology market is expected to overreach 5.3 trillion U.S. dollars. Only in America there are more than 500 000 IT and Software companies. Leading the world is Microsoft, followed by Apple and Nvidia. While Microsoft and Apple are in our every day life for the time I can remember my existence, Nvidia has gained the attention seriously in the past few months and was named the world’s most valuable company. I personally have invested in the company and decided to dive in the history and have a detailed look at its products.

NVIDIA is a technology company that designs and manufactures graphics processing units (GPUs) and system-on-a-chip units (SoCs) for the gaming, professional visualization, data center, and automotive markets. A system on a chip (SoC) is not the same as a CPU. While a CPU is a single component that handles the processing of data in a computer or electronic system, a SoC is an integrated circuit that contains many components of a computer system, including the CPU, on a single chip.

In summary, while a CPU is a part of a SoC, a SoC is a more comprehensive and energy-saving package that includes the CPU and other necessary components for a computer system to function. Here are one of NVIDIA's main CPU products: The Grace CPU Superchip.

The NVIDIA Grace Hopper Superchip is a groundbreaking processor specifically engineered for massive-scale AI and high-performance computing (HPC) applications. It has been meticulously developed to deliver exceptional performance in handling vast quantities of data, generating intelligent insights with unparalleled energy efficiency. This advanced technology is tailored for modern data centers that manage a wide array of workloads, including AI, data analytics, hyperscale cloud computing, and high-performance computing.

The NVIDIA Grace Hopper Superchip offers a remarkable 2X performance per watt, 2X the packaging density, and the highest memory bandwidth compared to today's top-tier servers. This enables it to meet the most demanding data center requirements. By harnessing the power of this innovative processor, data centers can process and analyze massive amounts of data with unprecedented speed and efficiency, ushering in a new era of computing capabilities.

More than ever we have rise of electric vehicles. More than that, Tesla has transformed itself into a company of the future, engineering the first Optimus robot and together with the gigantic factories and solar mega packs, is expected to reach the trillion dollars world as well anytime soon.

Tesla AI has developed few main different categories: The humanoid robot Optimus 1 and 2, the Dojo System, which is a super computer designed for vision video processing used for training machine learning models to improve the Self-Driving system. Neural Networks use input-output data and make intelligent algorithms take decisions.   

Tesla’s neural network per-camera networks analyze raw images to perform semantic segmentation, object detection and monocular depth estimation. The birds-eye-view networks take video from all cameras to output the road layout, static infrastructure and 3D objects directly in the top-down view. These networks learn from the most complicated and diverse scenarios in the world, iteratively sourced from millions of vehicles in real time.

The Tesla Bot was firstly introduced in 2021, followed by the enhanced Bot 2 developed and introduced in 2023. The Tesla Bot is a step towards a human-robots future which is expected to boost economic growth tremendously, possibly through enhancing production and human advancement. As cars and other automatic transport tools has brought a tremendous change in the world, the humanoid Optimus by Tesla is expected to accelerate the Earth’s potential.

Cost estimation is between 10 and 20 000 U.S dollars. This will come with the freedom to choose between different options according to customer preference. There will be mass production to make this estimated low cost of the robot. Unlike the human, Optimus has no need to rest, able to operate after just few minutes of super charging after battery depletion.

Until that moment the robot was trained to 30% walk speed boost forward, fully move its thumbs with 22 DoF(Degree of Freedom) compared to humans 27 DoF, a 2-DoF actuated neck coordination and do boring tasks. Can you imagine what the Tesla factories look like already? Further tasks will include assembly operations, playing music intruments, handling complex kids care, teaching. Everyone will be able to remotely control the performance of the tasks.  To become fully cooperative, the Bot will continue to be fully trained with the natural language model algorithms, having its own brain interface neural network.

These are the technologies and AI that make both Tesla and Nvidia progress the world.

Updates as of 2025 include start of production and further enhancements. Here is an overview:

Optimus Updates

• Production Milestone: Tesla has begun production of the Optimus robot, with the first unit reportedly produced at a new production line in Fremont. This marks a significant step from prototype development to initial manufacturing.

• Enhanced Hand and Forearm: A new Optimus hand with 22 degrees of freedom (up from 11) and an improved forearm design are now in production. This upgrade enhances the robot’s dexterity, allowing for more complex and precise movements, which could expand its capability to perform a wider range of tasks.

• Sophistication Claims: Elon Musk has described Optimus as the “most sophisticated humanoid robot on Earth,” highlighting its advancements in AI, electric motors, batteries, and manufacturing scalability. While specific new capabilities weren’t detailed beyond the hand upgrade, this suggests ongoing refinement in its autonomous functionality, such as navigation and task execution, leveraging Tesla’s automotive AI expertise.

• Future Projections: Musk has reiterated ambitious goals, including producing “several thousand” Optimus units by the end of 2025, with plans to scale to millions annually in the future. While these are forward-looking statements, they indicate active development and optimism about production ramp-up.

Dojo Supercomputer Updates

• Operational Status: The first-generation Dojo supercomputer is fully operational at Tesla’s Gigafactory New York and Palo Alto facilities. It is currently handling AI training workloads, contributing approximately 5% to 10% of Tesla’s total AI training capacity, with expectations of increasing its role.

• Next-Generation Plans: Tesla is planning a more advanced version of Dojo, projected to be “ten times” more powerful than the current iteration. While specifics on its architecture or timeline weren’t detailed, this suggests significant enhancements in computational power, likely aimed at accelerating neural network training for Full Self-Driving (FSD) and Optimus.

• Cortex Supercluster: Alongside Dojo, Tesla is developing a new AI training supercluster called Cortex at its Austin headquarters. Described as a “giant” system, it will incorporate around 100,000 Nvidia H100/H200 GPUs and massive storage for video training, indicating a dual-track approach to bolster AI compute capacity. This complements rather than replaces Dojo, focusing on broader AI tasks.

• Training Contribution: Dojo’s current role in training (up to 10%) shows it’s actively supporting Tesla’s AI efforts, though it’s still a smaller fraction compared to Nvidia-based systems. The planned Dojo 1.5, 2, and beyond versions hint at a long-term strategy to reduce reliance on external hardware like Nvidia GPUs.

These updates reflect Tesla’s ongoing efforts to integrate Optimus into its operations and enhance Dojo’s role in AI development, particularly for autonomy in vehicles and robotics. However, some claims—like exact production numbers or the full scope of Dojo’s next-gen capabilities—remain speculative or forward-looking, as Tesla often shares ambitious targets that evolve over time. For the most current details beyond March 30, 2025, further announcements from Tesla would be needed.