Tesla admits HW3 isn't enough for FSD, will set up micro factories to retrofit older models

During an earnings call for Q1 2026, Tesla's CEO, Elon Musk, revealed that the company is planning to upgrade to its self-driving chip. The name of the chip is undecided, but it will likely be AI4.1 or AI4 Plus. The new chip will double the RAM from 16GB to 32GB per chip, and since the system uses a dual-chip design, the total system memory will jump to 64GB.

Elon Musk finally admitted that vehicles equipped with the HW3 onboard computer won't be able to support unsupervised FSD after all, even though the company promised its customers that the HW3 computer inside Tesla vehicles sold between 2019 and 2023 is perfectly capable of handling the computational demand for FSD.

The next-generation AI5 chip is still being developed and won't go into production anytime soon. Besides, Elon Musk assures that AI4 and AI4 Plus will be more than enough for FSD, so AI5 will likely go into the Optimus robots and data centers instead.

Interestingly enough, an AI4.5 is already being sold in 2026 Model Y vehicles, coming out of the Fremont plant. That version of the hardware uses a three-chip design. So, a bit confusing, but the upgrade goes as follows: AI4 > AI4.5 > AI4.1/AI4 Plus. The latter will hit the market in 2027, though.

So what about customers who have already bought an HW3 vehicle and opted for the full FSD package that costs up to $15,000?

Well, Tesla is planning a costly and not-so-elegant solution to the problem. It will offer Tesla HW3 owners a discounted trade-in for a newer vehicle with AI4 hardware, or retrofit their existing car with a newer chip and also newer cameras. But in order to do so, Tesla will have to invest in micro factories across the US that will handle the upgrade. This will prove to be a costly endeavor, but it's seemingly the only way to get out of the situation.

Notably, only users who bought the FSD package will be eligible for the upgrade. Those who only purchased an HW3-equipped Tesla will be left out of the program.

We wonder if the AI4 hardware will have the same fate as HW3 because as the neural network grows along with its complexity, the computing demand will surely rise in the future.

Via