How can a robotaxi feel “ready” on a convention floor while still looking nowhere near ready for daily life? CES has become a proving ground for mobility theatrics: sprawling booths, polished demos, and just enough real-world detail to make autonomy feel inevitable. The show’s scale makes it impossible to see everything, but one pattern stands out quickly autonomous driving is no longer a side exhibit. It is a central bet. That visibility also makes the gaps harder to ignore, because the most convincing robotaxi story is rarely about a single sensor suite or a flashier compute box. It is about whether the system can survive the unglamorous parts: fleet uptime, messy edge cases, and the human processes hiding behind “driverless.”
The sharpest contrast comes when prototypes and limited services are placed beside an operator that has spent years turning autonomy into something resembling routine. The difference is operational maturity: dispatch, remote support, cleaning, charging, maintenance, mapping updates, and incident response plus the quiet UX details that make a ride feel normal rather than experimental. That is why the industry’s perceived lead is increasingly tied to the service layer, not just the driving stack.
Profitability is the second reality check. Even a competent autonomous ride does not automatically pencil out. A simple per-vehicle model from a $286K loss per vehicle over its useful life highlights how quickly support costs dominate: labor outside the car, depot operations, and the continuous maintenance burden of sensors and compute. The uncomfortable twist is that personnel can remain the biggest expense even when nobody is sitting behind the wheel. In that framing, “scaling” is less about adding cities on a map and more about driving down interventions, compressing turnaround time, and standardizing the fleet so each additional vehicle does not demand another shadow team to keep it moving.
That shadow team is also why teleoperation keeps resurfacing, despite the stigma. Remote assistance is often treated like an embarrassment, but the more practical view is that it is a bridge and, in some designs, a permanent layer. Teleoperation brings hard constraints: latency, variable networks, and cybersecurity exposure. As Roboauto CEO Jakub Juza put it, “Wireless networks are the oxygen for teleoperation technology.” He also drew a line between continuous low-speed control and selective high-speed intervention, noting that “the key variable is risk versus speed.”
Underneath all of this sits the trust problem: AI that is powerful, non-deterministic, and difficult to validate like traditional software. The shift toward software-defined vehicles is accelerating, with research cited by UL indicating that nearly three quarters of vehicles coming off assembly lines in the next decade could be software-defined and AI-powered. That raises the bar for safety arguments that go beyond “it passed our tests,” especially when standards like SOTIF (ISO/PAS 21448) exist specifically to address hazards from system limitations and edge cases rather than component failure.
CES can make robotaxis feel like a close race. Outside the booths, the separation comes from who has already built the boring machinery of autonomy and who is still auditioning for it.