Tesla is pitching FSD v15 as a transformative "step-change" in autonomous driving capability, but the company's track record suggests skepticism is warranted. During a recent meeting at its Fremont factory, Tesla told JPMorgan that version 15 of its Full Self-Driving software marks a major leap forward, built on seven core technologies with about 40% already running in the robotaxi fleet in Austin. However, Tesla made nearly identical claims about FSD v14, which shipped in October 2025, and v14.3, which Elon Musk called "the last piece of the puzzle" in April.

Tesla

Why Does Tesla Keep Promising the Same Breakthrough?

The pattern reveals a consistent gap between Tesla's investor messaging and its actual delivery. Since 2016, Tesla has been selling Full Self-Driving as unsupervised autonomous capability on vehicles customers own. Yet after a decade, customers are still required to supervise their cars. Every major version update has been framed as the definitive solution, only to be followed by another version with the same promise.

Tesla's current hardware, called HW4 or AI4, is being presented as sufficient to run v15 and enable unsupervised driving. This echoes Tesla's previous claims about HW3, the prior generation computer. For years, Tesla promised HW3 could handle unsupervised FSD before quietly walking back that commitment. Today, HW3 vehicles receive only a stripped-down "Lite" version of FSD v14, and the company has reported rising hardware failures tied to that software.

What's the Hardware Roadmap Actually Showing?

Tesla's chip development timeline suggests the company continues to shift its goalposts for which hardware will finally deliver unsupervised driving. The company delayed its next-generation AI5 chip to mid-2027 and confirmed that the Cybercab robotaxi will launch on AI4 or the newer AI4.5 variant, which offers roughly 10% more computing power and about twice the memory. This pattern of delaying the next chip while claiming the current one is sufficient has repeated multiple times in Tesla's history.

How to Evaluate Tesla's Autonomous Driving Claims

• Track Record Analysis: Compare current promises against previous versions. Tesla has called FSD v13, v14, and now v15 each a "step-change," yet customers still supervise their vehicles after years of ownership.
• Real-World Deployment Scale: Examine actual usage data rather than capability claims. Tesla's Robotaxi service in Austin has logged only 380,000 unsupervised miles since launching more than a year ago, despite the company's confidence in scaling.
• Hardware Sufficiency Claims: Scrutinize assertions that current hardware is adequate. Tesla made identical claims about HW3 before requiring hardware upgrades, suggesting AI4 may face similar limitations.

The robotaxi program itself illustrates the gap between confidence and execution. Tesla told JPMorgan it is deliberately holding back further Model Y robotaxi conversions because of confidence in the Cybercab's near-term scaling potential, using an "unboxed" manufacturing process and targeting a long-run cost of around $0.30 per mile. Yet the actual driverless service remains limited to a handful of cars more than a year after launch, with frequent issues reported despite extremely low mileage.

What About the Optimus Robot Timeline?

Tesla's humanoid robot program faces similar credibility challenges. The company told JPMorgan that the Optimus Gen 3 design is finalized, the supply chain is "essentially locked in," and a production line is being installed in the old Model S/X facility in Fremont. External commercial sales could begin "as early as" the second half of 2027, with long-term capacity ambitions of about 1 million units at Fremont and 10 million in Texas.

However, two significant factors undercut this timeline. Tesla has repeatedly delayed the Gen 3 reveal, and Musk himself admitted that no Optimus robots are currently doing useful work at Tesla, despite previously claiming the company would have 5,000 robots in operation. The pitch essentially amounts to: a robot Tesla hasn't publicly demonstrated, that cannot yet perform tasks inside its own factory, will be available for purchase by outside customers in less than two years.

The analyst delivering this message also matters. JPMorgan replaced Ryan Brinkman, one of Wall Street's longest-running Tesla skeptics, with Rajat Gupta, who upgraded the stock to Neutral and set a price target around $445. Gupta came to Tesla coverage from a bullish call on Carvana, and the note largely relays Tesla's own guidance from the Fremont visit without substantial pushback.

For investors and customers evaluating Tesla's autonomous driving promises, the pattern is instructive. Every claim about FSD v15 being a breakthrough, HW4 handling unsupervised driving, the Cybercab scaling into a profitable service, or Optimus launching in 2027 mirrors promises made about previous versions and hardware generations. The meaningful test will come when a customer vehicle actually demonstrates unsupervised FSD capability in real-world conditions, and Tesla accepts legal responsibility for its performance. Until then, these remain claims awaiting evidence.