Tesla (TSLA) CEO Elon Musk had a long meeting with his chip design team this past weekend. According to Musk, it was related to a design review of the upcoming Tesla AI5 (HW5) chip.
“This is going to be an epic chip,” Musk expressed his joy after reviewing the design of the Tesla AI5 processor. “And AI6 to follow has a shot at being the best by AI chip by far,” he added.
The automaker aims to use this next-generation supercomputer chip for its neural-net inference & training for Autopilot Full Self-Driving (FSD) and Tesla Optimus robot’s vision.
Elon Musk’s short note from the meeting suggests that the design for the Tesla AI5 artificial intelligence chip is close to completion.
Taiwan Semiconductor Manufacturing Company Limited (TSMC) has already won the contract for the manufacturing of the Tesla AI5 chips. TSMC will first produce these chips in Taiwan and in the next phase, will move manufacturing to Arizona, United States.
One of the major changes from AI3 (HW3) and AI4 (HW4) circuit boards to AI5 is the use of a single chip. AI3 and AI4 motherboards used in Tesla cars had dual chips installed.
According to Elon Musk, with AI5 and AI6 chips, Tesla is moving toward a single-chip architecture. This also shows the company’s confidence in making reliable supercomputer processors because, in previous versions, the second chip was basically used for redundancy.
“Switching from doing 2 chip architectures to 1 means all our silicon talent is focused on making 1 incredible chip. No-brainer in retrospect,” Elon Musk wrote on X.
“I think AI5 will probably be the best inference chip of any kind for models below ~250B params. By far the lowest cost silicon and best performance/Watt,” he further stated.
Retirement of the Dojo Supercomputer Cluster
Elon Musk, his Tesla silicon, and neural net architecture teams last month decided to retire the Dojo supercomputer cluster. Although it was clearly headed for the next phase (Dojo 2) later this year, Musk directed all focus and resources to AI5 and AI6 chip design and development.
“Once it became clear that all paths converged to AI6, I had to shut down Dojo and make some tough personnel choices, as Dojo 2 was now an evolutionary dead end,” Musk said on X.
“Dojo 3 arguably lives on in the form of a large number of AI6 SoCs on a single board,” he added.
For Full Self-Driving (FSD) machine learning (ML), Tesla is already running multiple neural net data centers. Tesla’s Cortex supercluster at Giga Texas is one of the largest neural net ML facilities that is currently using Nvidia H100 and H200 GPUs for the purpose.
However, as Tesla is progressing with in-house built chips/processors, the automaker will completely transition to AI5 and AI6 chips in the future.
Tesla awarded Samsung a $16.5 billion contract in July for producing Tesla AI6 chips starting in 2026. Samsung is going to manufacture these chips at its semiconductor fabrication plant in Austin, Texas. This Samsung factory is just 40 40-minute drive from Tesla’s Giga Texas.
Musk talked further about the Samsung plant, he said:
Samsung agreed to allow Tesla to assist in maximizing manufacturing efficiency. This is a critical point, as I will walk the line personally to accelerate the pace of progress.
And the fab is conveniently located not far from my house
Stay tuned for constant Tesla updates, follow us on:
Google News | Flipboard | X (Twitter) | WhatsApp Channel | RSS (Feedly).
Related Tesla News
- Starship Booster 19 survives the cryo test, Flight 12 updates from Starbase
- Tesla files for additional trademarks for the Cybercab in some US states: Cybercar, Cybervehicle, Robotaxi
- SpaceX acquires xAI to build AI data centers in space, Tesla (TSLA) invests $2 billion in Musk’s xAI
- Watch the Cybertruck pull a Ford F-350 plus a large trailer from deep sand
- Tesla launches Model Y Standard AWD in the US with reasonable price, range, and better acceleration
- Tesla Cybercab butterfly door function spotted in the wild (video)
