A Launch That Deliberately Left Room for Chaos
Tesla's Cybercab launch offered an unusually open test of its robotaxi technology by allowing guests to choose destinations and ride through Austin rather than limiting them to a tightly controlled demonstration route. The geofenced rides could last roughly 45 minutes, giving attendees more freedom than a typical staged launch.
Host Phil argues that this willingness to let ordinary riders choose where the vehicles went was one of the event's most important signals about Tesla's confidence in the system.
He read the secrecy around the location as a calculated trade-off, no livestream meant no crowds blocking rides, no protesters jamming roads, and no safety exposure for a technology Tesla is still validating.
But he was blunt that the company created its own communications problem by leaving hundreds of thousands of people waiting for a livestream that never came, calling it a 'big unforced error' that Tesla could have fixed simply by setting expectations in advance.
The event also doubled as a celebration for roughly 250 employees, many of them young engineers who had spent a grueling 23 months converting the handbuilt prototypes shown at the We, Robot event into a production car. The guests who stayed past closing described a team photo, cheering, and a pride they said reflects thousands of people whose work is usually collapsed into the founder's name.
Two Data Points and a 17% Growth Rate
The event handed analysts a rare gift: an intermediate data point between earnings calls. Tesla said on stage it had reached 1 million unsupervised paid robo-taxi miles, confirmed by the app, up from the 360,000 miles reported around the July 22 earnings call.
Spread over roughly six weeks and a day, that works out to 17.05% week-over-week compounding growth, a figure the show's analyst emphasized is big but also off a very small base, where rapid growth is easy.
Extrapolating that rate, the analyst projected about 3.5 million miles by the October earnings call and over 14 million miles by year end, while stressing that this is arithmetic, not a forecast, and that the rate will likely slow as the base grows and supply constraints bite.
Translating miles into vehicles, a fleet of about 137 cars averaging 150 paid miles per day would support the current growth, rising to roughly 500 by the Q3 call and about 2, 000 by year end, or around 3, 000 if daily utilization drops to 100 miles.
The hard constraint is production. One analysis cited on the show put Cybercab capacity at a ceiling of 125, 000 units per year, or roughly 2, 400 per week and about 500 a day, a ceiling that continued 17% growth would hit sometime in April.
That ceiling is not fixed, the hosts cautioned, and it excludes Model Ys still joining the fleet, but it is the first real friction point between demand-side enthusiasm and factory throughput.
Starlink Moves Into the Car
Among the engineering displays at the event, the one that most excited the technically minded host was an exploded Starlink phased-array antenna destined for the Cybercab roof, a roughly one-foot-square unit he described as effectively a Starlink Mini. The layered display showed a weatherproof radome, stacked patch antennas, beam-former chips that electronically steer the beam toward satellites, and a housing designed to mount on the vehicle.
The practical payoff, he argued, is bandwidth and always-on connectivity: today's Teslas rely on paid 5G service for vital real-time data while holding back FSD uploads and software updates for home Wi-Fi.
With an integrated Starlink antenna wired to the car over Ethernet, those constraints loosen, in one anecdote, a software update downloaded mid-journey over a Starlink unit the owner had mounted in his Model 3, something that would otherwise have waited until he got home.
The longer-term implication goes beyond convenience. Because the car's cameras currently feed the FSD pipeline, richer connectivity could let fleet operations see what a car sees in real time and reroute the fleet away from congestion or danger, the hosts cited the 2026 World Cup crowds in downtown San Jose as the kind of situation an autonomous car cannot recognize today.
The antenna could even shrink further over time, since bit-packing and encoding gains have steadily increased bandwidth per antenna over the last two decades of RF technology.
Why Version 15 Needed September, Not June
Every Cybercab running today is on FSD version 15, and the show's engineering-minded guest argued the timing was no accident. His reconstruction: Ashok Elluswamy's team concluded they needed a bigger model (more parameters, capable of encoding thousands of planned reinforcement-learning use cases) and promised Elon Musk it could be built in record time.
That promise apparently held: version 15 reached the fleet by early September, roughly a quarter and a half to two quarters of concurrent work on model architecture, RL use cases, and performance optimization.
The harder half of the problem was making the larger model run on the car. Cybercabs carry AI4 processors (the host said he does not believe they have an upgraded 'AI4+' yet, which he expects in early 2027) so the team needed aggressive optimization to hit inference-cycle targets.
He connected that work to the 'version 14 light' release on hardware 3 cars as evidence of the investigative effort that paved the way for version 15 on AI4.
Validation added a third workstream: Cybercab is a new vehicle with no daily consumer-driving data stream to lean on, so it needed its own validation mileage beyond the 1 million paid miles to reach internal KPIs for matching Model Y performance.
The next expression of the same logic is a new active safety feature in which FSD supervised activates on the driver's behalf when an imminent collision is detected, a feature the hosts found intriguing but whose subscription requirements and liability treatment remain unclear.
Cheap Miles, Shifting Cities, and a Labor Fight
Both hosts flagged that the downstream effects on cities are genuinely unknown, cheaper, more fluid movement could fill restaurants or empty them, and robot-managed traffic should eventually beat human drivers at congestion. One thought experiment: if households give up cars and save, say, $5, 000 a year on transportation, most of that money flows into discretionary spending that benefits the wider economy, before Optimus someday pulls spending back toward the home by becoming the chef.
The more immediate fight is labor and regulation. The hosts aired a pointed exchange between commentators about whether Uber's network of partners can survive vertically integrated autonomy, with one guest arguing that Uber (once the disruptor of taxi drivers) is now positioning itself as the regulator-friendly protector of drivers, a irony he found hard to miss.
His own prescription is cybercab ownership by individuals, including former Uber drivers, as a 'win-win-win' for consumers, earners, and Tesla, and a way of spreading political risk across thousands of small fleet owners.
Beyond economics, the hosts closed on a human note: a blind speaker at a Nevada planning commission who described planning government-subsidized trips three days in advance, only to have Uber drivers pass her by because loading her wheelchair was a burden. A camera-only car, she said, is blind to her disability, and that is the point.
In the hosts' framing, the Cybercab is 'an object for freedom, ' and its success will be measured less in launch-night spectacle than in whether cars in the wild keep collecting passengers and revenue.