Lidar Is Paid For in Hours: Robotaxi Closes at 11 PM

A sensor's price reads in hours of service

A lidar left off the vehicle is a discount collected once, on the day the car rolls out of the factory.

The lost hours of service come back every night, for the whole life of the vehicle. On 3 October 2026 Tesla's Robotaxi account announced that the Austin service now runs until 11 PM. Elon Musk explained the limit with a line that went viral: the goal is to avoid hitting "grey kittens on grey asphalt" in the dark, as Electrek^[1] reports.

The right metric here is dollars per available vehicle-hour. On that scale a camera-only stack pays its own discount back with interest.

The consensus measures the bill of materials because it is a public number, easy and immediate. The number that decides revenue is a different one: the calendar of hours in which the fleet accepts rides.

Evening rides hold up the margins of any urban fleet, from the traditional taxi to the ride-hailing platforms. A platform that closes at 11 PM hands that band to its competitors. This is a regime change in sensor accounting.

Three dated points, a trajectory heading down

A curve needs three points, each with a value, a date and whoever measured it.

First point: June 2025, launch in Austin, the app accepted requests from 6 AM to midnight, that is an eighteen-hour window. Second point: until 2 October 2026 closing time sat at 10 PM, so sixteen hours. Third point: from 3 October 2026 closing time slides to 11 PM, seventeen hours. The three values come from Electrek's reporting of 3 October 2026.

Fifteen months after launch the window stays one hour shorter than on day one. The direction of the curve, over that period, points down.

Whoever reads the announcement sees an hour gained and applauds. Whoever reads the time series sees an hour lost and half recovered. 90% of analysts get the present right and get the pace of change wrong.

A fleet working seventeen hours instead of eighteen gives up roughly 6% of the operating day. That 6% weighs on the denominator of every fixed cost: depreciation, depot, insurance, software.

Why a camera loses the game in the dark

A camera is a passive sensor: it works with the light that reaches it and separates the object from the background by contrast. A grey animal on grey asphalt, at night, takes both of those away.

Lidar is an active sensor. It fires laser pulses, measures the return time and derives distance and shape in three dimensions. At midnight it works as it does at noon, because it produces its own light.

Radar does a similar job with radio waves and adds velocity. On a small target it yields little, in exchange it holds up through fog, dust and reverberation.

The grey cat is an almost comic example, and yet it describes a whole class of targets: pedestrians in dark clothing, low debris, stationary bicycles. The residual risk of that class decides the closing time.

Waymo's sixth-generation system carries thirteen cameras and four lidars. The redundancy costs in bill of materials and buys hours of service: it is a purchase, more than a ballast.

The cliff event: when the buyer is a city

The turning point arrives when the customer becomes a public administration.

On 1 October 2026 the Minneapolis city council brought robotaxi rules to a vote, with Waymo at the centre of the file, as MPR News^[2] and the Star Tribune^[3] report. In Pittsburgh the same company has begun testing in autonomous mode^[4].

On 1 October 2026 TechCrunch^[5] reported that robotaxi operators will face fines when they block emergency vehicles. A fine is an hourly cost dressed up as a rare event.

Cities already have the right vocabulary: they have used it for decades with buses and with taxis, where the licence imposes shifts and minimum coverage. Autonomous transport enters that vocabulary now.

The moment an ordinance writes "night coverage" into a concession, the discount on the bill of materials evaporates. From that day the buyer looks at availability first, then at the price of the vehicle.

Three categories that change shape by 2029

Three categories reach 2029 in a shape different from today's.

• Automotive lidar and radar suppliers
• Fleet insurers
• Procurement offices that sign multi-year contracts

Active sensor suppliers stop selling a part and sell guaranteed hours. The price per unit falls, the contract lengthens, the margin migrates to the service.

Fleet insurers price by sensor class, as they do today with automatic braking systems. A pricier policy for a passive stack turns the initial discount into an annual cost. The books record it for eight years.

Procurement offices change the line in the tender: in place of the vehicle price comes dollars per available vehicle-hour, with penalties tied to the time band. Anyone signing today on a camera-only fleet locks in for years a metric the market is about to abandon. It pays to ask for a sensor upgrade clause, before the request becomes obvious and expensive.

Where patient capital is looking

For a fund the uncomfortable bet sounds like this: the active sensor returns to centre stage precisely as the dominant narrative writes it off.

The mechanism is simple. The bottleneck migrates, and in this cycle it migrates from daytime perception to coverage of the difficult hours. Whoever prices yesterday's scarcity as permanent misses the turn.

For a head of technology today's move is an internal measurement: count your own availability hours by light class, in place of the model's average precision. The data already comes out of the fleet logs.

For whoever writes a three-year plan the point is a different one: a plan built on passive stacks assumes a world in which cities stay silent about hours. The October 2026 ordinances say the opposite.

My position, and the evidence that would change it

This desk's position fits in one line: the bill-of-materials discount on a camera-only stack is deferred debt, and it is paid in lost hours of service. A sensor is judged by the cost per hour of service it enables, never by the price of the part.

The supporting evidence comes from the news: fifteen months after launch, the Austin window stays shorter than on day one, and the stated reason concerns contrast in the dark.

The contrary evidence would take this form: Tesla reopens the band until midnight, in more cities, with the same sensor set. It would also need a rate of contacts with pedestrians and animals equal to that of lidar-equipped fleets. In that case the thesis falls.

Confidence high on the physics, medium on the industrial calendar. Light obeys stable rules, corporate plans obey quarters.

Forecast, horizon, kill signal

Forecast: as of 30 June 2027 Tesla's Robotaxi service in Austin still closes before midnight.

Confidence: 70. Horizon: 268 days, expiring on 30 June 2027. Kill signal: published hours running until midnight or beyond, for seven days in a row, before that date.

An honest market indicator for this thesis is missing: one hour of service in one city moves zero on the price of any listed stock.