Frustration has been building across the Tesla community over the lack of clear visibility into Robotaxi scaling plans. Elon had promised an epic year 2026 and a ridiculous 2027. It is now clear that this will be completely missed.
Retail shareholders have pressed the company with pointed questions on Say Technologies ahead of recent earnings calls. The previous quarter’s discussion felt vague and imprecise. This time, however, Tesla’s senior management appeared to have taken the feedback on board. Ashok arrived prepared with concrete numbers.
With actual figures on the table, basic math is enough to start unpacking what they imply. Let’s dive in.
Part 1: Do Ashok’s Numbers Hold Up?
Ashok communicated two important figures: Robotaxi has driven 380,000 unsupervised miles since the beginning of the year, and the week-over-week growth rate of those unsupervised miles has been “more than double-digit.”
What does “more than double-digit” actually mean? At a minimum it is 11%. A realistic central estimate is around 14–15%. Anything much above 17% would likely have been described differently. We therefore examine three scenarios:
- Pessimistic: 11% weekly growth
- Base case: 14% weekly growth
- Optimistic: 17% weekly growth
We take calendar week 1 of 2026 as the starting point. The end of 2025 consisted of little more than a handful of rides (including the now-famous “Feline Self Driving” supervised-by-cat phase).
The cumulative miles after k weeks of geometric growth form a standard geometric series. If m₁ is the miles driven in week 1 and the weekly growth factor is r=1+x/100, then the total miles after k weeks is:
Using the known total of 380,000 miles by calendar week 30, we can solve for the three possible values of m₁. From there it is straightforward to reconstruct the weekly miles for every earlier week.
The Robotaxi tracker (while it was still reliable) recorded 2 unsupervised vehicles in CW4 and 38 in CW19. Comparing those fleet counts with the modeled miles in the same weeks gives the implied daily mileage per vehicle:
The CW4 numbers are noisy because of the tiny sample. The CW19 figures are far more informative: all three growth scenarios converge on roughly 37–47 miles per vehicle per day. A working baseline of 50 miles per vehicle per day therefore looks reasonable for the current stage of operations.
Using that 50-mile assumption, we can back out an implied unsupervised fleet size:
The horizontal line marks the 102 DMV-registered units known by mid-July. In all three scenarios the modeled fleet approaches that level at roughly the same moment Tesla was observed adding about 75 new vehicles. This is a useful consistency check on Ashok’s double-digit growth claim.
That said, the extrapolated fleet-size numbers themselves should be taken with a grain of salt. Tesla can also grow total unsupervised miles by raising daily utilization per vehicle (moving from today’s ~50 miles toward 100+ as the network densifies). The graph is therefore best read as a cross-check on the growth rate, not as a precise forecast of how many cars will be on the road.
With that validation in hand, we can turn to longer-term projections.
Part 2: Projecting Miles and Fleet Growth Through 2027
Now that Ashok’s numbers appear consistent with the available data, we can turn to forward-looking projections. Ashok stated that the current double-digit week-over-week growth is expected to continue through the rest of 2026. That gives us a usable window for modeling.
End-of-2026 fleet size
While unsupervised miles are what ultimately drive revenue, fleet size remains a useful headline metric in the near term — especially while revenue is still negligible. It is also a practical proxy for whether logistics, charging, maintenance, and regulatory issues are being solved at scale.
Continuing to assume only 50 miles per vehicle per day through year-end is unrealistic. As the network densifies, daily utilization should rise. For the end-of-2026 projection we therefore adopt a baseline of 100 miles per vehicle per day.
Under that assumption the three growth scenarios produce the following unsupervised fleet sizes by the end of 2026:
- 11% → roughly 560 vehicles
- 14% → roughly 1,200 vehicles
- 17% → roughly 2,500 vehicles
The most probable outcome sits around 1,000 units. That is materially lower than many early expectations. Double-digit weekly compounding is powerful, but when the starting base is small it simply takes time before the numbers become large.
Cybercab production line utilization
A second, related question is whether the projected growth is consistent with sensible use of the Cybercab production line. Running an expensive new line at very low utilization would represent poorly deployed capital. To test this, we adopt a high-utilization assumption of 200 miles per vehicle per day — toward the upper end of what a well-run robotaxi service can sustain.
We then ask how many new vehicles must be added each week to support the three growth trajectories, and when those additions would hit the practical capacity of the Cybercab line. Tesla has guided to approximately 125,000 Cybercab units per year in the Q2 2026 shareholder deck. We therefore use a baseline capacity of 2,400 vehicles per week.
The projection shows that under all growth cases the line approaches full capacity before the end of 2027 — roughly 18 months after start of production. Reaching near-full utilization on that timeframe can reasonably be described as well-planned and well-executed. The short-term numbers may still look modest, but once the exponential curve steepens, a rapid ramp in Cybercab output becomes visible from the middle of 2027 onward.
Part 3: What This Means for EPS
We can now turn to the question that matters most: revenue and earnings.
First we project weekly unsupervised miles under the three growth trajectories, keeping the current compounding rates running:
To convert those miles into EPS we apply a set of straightforward assumptions:
- Only 50% of unsupervised miles are paid (the rest being empty repositioning, charging movements, etc.)
- Revenue of $1.50 per paid mile (consistent with current Robotaxi pricing observed in the tracker)
- 50% gross margin
- 25% tax rate
- 3.5 billion diluted shares
The resulting quarterly EPS contribution from Robotaxi alone is shown below:
For context, Tesla’s recent quarterly EPS has been in the low-to-mid $0.30s. Against that backdrop the table is unambiguous on two points:
- 2026 contributes essentially nothing. Even the optimistic case remains well under one cent for the full year.
- An inflection appears around the middle of 2027. By the fourth quarter of 2027, Robotaxi is a meaningful profit contributor in every scenario — and in the higher-growth cases it becomes large.
This timeline lines up with comments Elon made in the prior quarter: Robotaxi would not be a meaningful source of income in 2026, but would become meaningful the following year.
Returning to the broader framing, 2026 was not the epic year that had been promised. The data we have, however, still points to 2027 as the year the exponential becomes visible in the financials — and to 2028 as the year that could finally look “ridiculous.”
Conclusion
Tesla has finally provided quantitative data on the pace of the Robotaxi rollout. The figures Ashok shared fit tightly with the fleet growth observed so far, and they allow us to make grounded forecasts.
The numbers will remain small this year. That will disappoint many. Yet it is worth remembering what exponential growth actually looks like: painfully slow at the start, then suddenly decisive. The data points to the long-awaited acceleration becoming visible toward the end of summer 2027 and unmistakable by year-end.
Critics will continue to argue that Tesla perpetually pushes deadlines and never delivers. That critique was valid in 2020 and 2021. FSD was not ready; any test drive made the gap obvious. The situation today is different. FSD is in customers’ cars. Users report thousands of miles between interventions. Safety-related events are rare. Robotaxi itself has accumulated 380,000 unsupervised miles without a single at-fault incident. This is hard data, not projection.
Applying an argument that made sense five years ago to the present reality requires ignoring what has actually changed. The day the thesis breaks is the day clear technical failure appears, or the day a competitor demonstrably executes faster and better. Apart from XPeng — still well behind and barred from the U.S. market — no other company is combining the necessary ingredients at scale: custom inference hardware, a closed data flywheel, bit-exact inference in the training loop, and a dedicated supercluster.
The competitive landscape remains empty. Tesla is executing. The outcome is inevitable.
