InfrastructureMay 1, 20267 min read

Robotaxi infrastructure 101: the four pillars

Most conversations about autonomous vehicles focus on the car. The harder conversation is about the infrastructure that has to exist before the car is useful at scale. Robotaxi infrastructure rests on four pillars. Pull any one out and the system collapses.

Interior of an Autzu Giga Hub showing the four pillars of robotaxi infrastructure operating in one footprint — energy, robotics, software and real estate.

Pillar 1 — Energy

A fleet of 250 robotaxis driving 200 miles a day consumes roughly 1.4 megawatt-hours of electricity per day per 10 vehicles, peaking sharply in the early morning and late afternoon. Without on-site energy management, every fleet either trips utility demand charges into the millions per year or starves its vehicles of charge during peak revenue hours.

Real fleets handle this with a mix of on-site stationary batteries, dynamic load-balancing, time-of-use pricing logic, and (in some markets) solar canopies. The goal is the same: smooth the demand curve so the utility bill behaves and the charging plaza always has electrons when the dispatcher needs them.

Failure mode without it: vehicles queue, drivers wait, ride availability craters at peak demand, and the unit economics fall apart.

Pillar 2 — Robotics

No human is going to clean a vomit incident at 3am, replace a cracked LiDAR cover at scale, or recalibrate 40 sensors in 60 minutes. The hub has to do it. Robotic wash arches, automated inspection rigs, gantry-mounted sensor cleaners, and turntable-based calibration lanes are not nice-to-haves — they are the only way to hit the cycle time the dispatch demand curve requires.

The benchmark we hold ourselves to: median vehicle turnaround under four minutes, p95 under eight. Anything slower and the fleet is throughput-bound at the hub instead of demand-bound on the road.

Failure mode without it: human labour costs and human error swamp the model. Cycle time blows out. Vehicles spend more time in the hub than on the road.

Pillar 3 — Software

The four operational questions every fleet operator answers every minute are: which vehicle, which task, which charger, which dispatch slot? Doing that across thousands of vehicles requires a real-time platform that fuses telemetry, energy state, asset health, sensor calibration status, and trip demand into a single decision surface.

This is not an MVP problem. Every minor optimisation — a 30-second improvement on charger handoff, a 2% improvement on dispatch packing — compounds into millions of dollars of fleet revenue per year.

Failure mode without it: the operator runs the fleet on spreadsheets and Slack. Utilisation stays under 50%. Margin disappears.

Pillar 4 — Real estate

AV hubs are dense. A fully-equipped Giga Hub needs roughly 30 sq ft of utility-fed plaza per vehicle in the steady-state fleet, plus service bays, calibration lanes, and dispatch staging. In the cities AVs serve, that real estate is contested, expensive, and often zoned against fleet operations.

Operators that solve this early — via long-term ground leases, brownfield site conversions, and joint ventures with municipalities — pull ahead. Those that don't end up running short pilots in the suburbs and never reach the density that makes the unit economics work.

Failure mode without it: the fleet is structurally too small to be profitable in the market it serves.

Why operators partner instead of building this themselves

Each of these four pillars is a multi-year, multi-million-dollar build. AV operators are already absorbing the cost of building the autonomy stack itself. Outsourcing the infrastructure layer to a specialist is how rideshare networks scaled in the 2010s; it is how robotaxis will scale in the 2020s.

References & further reading

  • BloombergNEF — Electric Vehicle Outlook 2025bnef.com
  • IEA — Global EV Outlook 2024iea.org
  • RethinkX — Rethinking Transportation 2020–2030rethinkx.com
  • McKinsey — Mobility's second great inflection pointmckinsey.com

Continue reading