Korea ML.co site reports that Tesla AI5 chip will have 2000~2500TOPS performance as early as 2026. (Tesla calls it HW4 and AI4 interchangeably from HW4). To this end, Tesla is working with Samsung Electronics and TSMC as contract manufacturers for AI5. Mass production is likely to begin next year. Tesla vehicles are expected to start being equipped with HW5 (AI5) in the second half of 2025, likely around December.
During a past earnings call, Elon claimed AI5 could be as much as 10 times more capable than HW4, which would imply 5,000 TOPS. There could be two AI5 chips for each Tesla car and one for each Teslabot. A combination of chips in packages would align the previous statement with the new specifications. The previous claim from 2024 could be adjusted based upon actual mass production specifications.
Tesla and xAI are invested more than $10 billion (KRW 13.7 trillion) in AI last year alone. Tesla installed 50,000 NVIDIA H100 GPUs in Cortex in the early days, and they plan to secure up to 350,000 more in the future.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Hm… Elon said that AI5 would be a “B200-class” chip. But B200 has 4500 TOPS at 8 bits and a whopping 9000 TOPS at 4 bits. The 2000 to 5000 tops, is that at 8 bits? If yes, then the AI5 is actually a B200 class chip.
Lets hope it takes less than the almost 3 years going from AI3 to AI4 when making the jump from AI5 to AI6. Nvidia can make one processor per year, at least.
I wonder how many cars will have to be outfitted with this (or even better), for free, just to fulfil promises about FSD. Besides raw compute, memory seems to be a showstopper. All automotive designs so far, including Tesla, have far too little memory to be able to do what they are supposed to.
How much memory in HW4?
Robotaxi in Austin is using Stock model Ys out of the GF so they have HW4. Tesla will have to update HW3 FSD buyers with HW4 at their expense, but HW5 won’t be part of it.
I wonder how hard it will be to replace my AI4 computer in my 2024 model why with an AI5 computer?
It’s actually very easy. Just go to Tesla’s website, and you can order a new car, it will come with it. Boom, i just helped you out.
LOL
2,000 and 5,000 TOPS is equivalent to AMD MI400 and MI500. AI chip performance roughly triples every two years. Moore’s Law doubled every two years. Power efficiency or tokens per Watt is the competitive factor. A purpose built ASIC is usually more power efficient than a general purpose device.