Tesla AI5 and AI6 Memory Cut for Optimus Explained: 72GB, Then 96GB, and 144GB (What Musk Actually Said)

TL;DR: On 1 Oct 2026 (US time), Elon Musk wrote on X that Tesla cut the memory on its AI5 chip in half, to 72GB of LPDDR5, and on its AI6 chip by a third, to 144GB of LPDDR6, as "the only way to get enough volume for Optimus production." The next day he wrote that AI5 would be nudged up to 96GB. So the latest figures from Musk himself are AI5: 96GB LPDDR5 and AI6: 144GB LPDDR6. The claim that this will barely hurt Optimus performance is Musk's own opinion, not a measured result.

What was said, and when

All times converted to Korea time (KST) from the post timestamps we read on X.

• 1 Oct 2026, 19:51 UTC (2 Oct, 04:51 KST): Musk, in a reply on X: "We cut our RAM in half for the Tesla AI5 chip (now 72GB of LP5) and 1/3 for AI6 (now 144GB of LP6). This was the only way to get enough volume for Optimus production and greatly reduces cost. As it turns out, we think this will have a negligible effect on Optimus performance, as memory bandwidth is a bigger limiting factor than total memory storage (bandwidth was held constant)."
• 2 Oct 2026, 07:56 UTC (2 Oct, 16:56 KST): Musk, replying to the same thread: "Actually, we decided to nudge AI5 up a little to 96GB, as Tesla would otherwise be the only company using the min RAM version of LP5."
• 1-2 Oct 2026: Benzinga reported the first post. Its article (dated 1 Oct, US evening) didn't include the 96GB follow-up, so a reader of that piece alone would think AI5 is 72GB.

Why Tesla says it did this

Benzinga ties the post to Micron's earnings call the day before, on which Micron CEO Sanjay Mehrotra said humanoid robots are expected to need more than 200GB of memory and multiple terabytes of storage each, similar to autonomous vehicles. Benzinga also says Tesla has reportedly placed a first large-scale component order for about 5,000 Optimus units, and that Tesla aims to build 1 million Optimus a year eventually. The second of those is a "reported" item, not something we confirmed.

Musk's own reasons, from his posts:

1. Volume. The cuts were the only way to get enough memory to build Optimus in volume.
2. Cost. He says it greatly reduces cost.
3. Second source (96GB bump). Using the smallest LPDDR5 configuration would have left Tesla as the only buyer of that variant.

What "bandwidth vs capacity" means

Capacity is how much data the memory can hold. Bandwidth is how fast data moves in and out. Musk argues that for a robot running AI models in real time, speed of access limits performance more than total size, and that bandwidth was kept the same. That is plausible engineering logic, but we found no benchmark, Tesla document or independent test that confirms the effect on Optimus. Treat "negligible" as a company claim.

What we can and can't say

• Confirmed from Musk's X posts: the 72GB and 144GB figures, the stated reasons, and the 96GB nudge for AI5.
• Not confirmed: whether 96GB is final, how many Optimus units Tesla will build, when AI5 and AI6 ship, or how much money this saves. Tesla hasn't published a spec sheet that we could find.
• A secondary site claims more. One smaller outlet, AIReport.net, says the earlier AI5 plan was 144GB and that Tesla's investor materials mentioned 9x memory capacity versus AI4. We could not verify those details from a primary source, so we don't rely on them.
• Where the chips are used: Musk's posts frame this around Optimus. We have not seen a source in these posts saying what it means for Tesla cars, so we don't claim anything on that.

Practical takeaways

If you follow Tesla as a customer: Nothing here changes what you can buy or use today. These chips are for future hardware, and the posts don't set a date.

If you follow Optimus or AI hardware: Treat spec numbers from social posts as moving targets. This one changed within about 12 hours.

If you invest: This article is not investment advice. Memory supply and cost are real constraints for any company planning robots in large numbers, as Micron's comments show, but a CEO's post is not a financial disclosure. Look for Tesla's filings and earnings materials.

How to check updates yourself: Read the original posts on X, and watch for Tesla's next earnings call or a spec release. If Tesla changes the numbers again, the latest official statement wins.

Sources: Elon Musk's posts on X (1 Oct 2026, ID 2105747471045370250, and 2 Oct 2026, ID 2105930010205024419), read through the X API; Benzinga, "Elon Musk Says Cutting Optimus Memory Was Tesla's 'Only Way' to Scale..." (1 Oct 2026); AIReport.NET (secondary, used only for context). Read 16 Oct 2026. Not investment advice.

---

← Back to all posts