Tesla hosted JPMorgan analysts at its Fremont factory and sent them away with an updated set of timelines for Full Self-Driving, Cybercab and Optimus. The bank's note is now public, and it is worth reading twice: once for what Tesla is claiming, and once for how much of it Tesla has claimed before.
Nothing here is an official Tesla announcement. It is a bank's summary of what Tesla told it, which is a lower standard of evidence than a release note.
What Tesla says about FSD v15
Tesla describes v15 as a "step-change" built on seven "core technologies", roughly 40% of which are said to be running in the Austin robotaxi fleet already. Early feedback is "encouraging".
The caution is that v13 and v14 were both pitched in similar language. Musk called v14.3 "the last piece of the puzzle" as recently as April. A step-change that arrives every version is a release cadence, not a step-change.
The hardware claim is the one to watch
| Claim | Status |
|---|---|
| AI4 sufficient for unsupervised FSD | Tesla's assertion, unproven in a customer car |
| AI4.5: ~10% more compute, 2× memory | Tesla's assertion, rolling out now |
| HW3 unsupervised capability | abandoned; HW3 cars get FSD v14 Lite |
This claim has the longest paper trail behind it. Tesla said for years that HW3 was sufficient for unsupervised autonomy, then stopped saying it, and HW3 cars now receive a reduced feature set while failure rates on those units rise. An owner deciding what to pay for FSD today is being asked to accept the same category of promise about AI4 that did not hold for HW3 — and a builder has already shown that retrofitting AI4 into an older Model 3 is physically possible while Tesla still does not offer it.
Cybercab, and why Model Y robotaxis are being held back
Tesla expects Cybercab to scale on its "unboxed" assembly process toward a long-run cost of $0.30 per mile, with fleet expansion accelerating from late 2026 into early 2027. It also told the bank that Cybercab is only the first robotaxi form factor, with further vehicle types planned.
On the Model Y, Tesla conceded that integration into the robotaxi fleet is deliberately slow, and gave a reason: Cybercab and v15 are what it intends to scale on, so pouring Model Ys into the fleet now would mean scaling the platform it plans to move off.
Set against that: after more than a year, Austin's driverless service runs on a handful of cars, and Tesla's cumulative total is around 380,000 unsupervised miles. Waymo is past 200 million and does roughly 500,000 paid rides a week.
Optimus
Tesla says the Gen 3 design is finalised, the supply chain "essentially locked in", and production begins in the old Model S and X space at Fremont. External sales could start "as early as" the second half of 2027, against capacity ambitions of one million units at Fremont and ten million in Texas.
The reveal has already slipped repeatedly, and Musk has acknowledged that no Optimus robot currently does useful work at Tesla, after previously talking about 5,000 units last year.
What this is worth to a European owner
Two concrete things. First, AI4.5 arriving now with more compute and memory means the hardware in a car bought this year is not the hardware Tesla intends to run unsupervised FSD on — directly relevant if you are weighing the FSD purchase in a European market where the feature is not yet approved. Second, the note contains no European regulatory timeline at all. Every date in it is a US date.
For the record on the analyst: JPMorgan replaced long-time bear Ryan Brinkman with Rajat Gupta, who upgraded Tesla to Neutral with a price target of about $445.