10,000 robotaxis land in Toronto. Where does the power come from?
Fleets arrive in months. Grid connections take years. And the years are longest exactly where the demand is.
Every number below is an assumption. Change any of them.
Underlined numbers are editable. Everything downstream recalculates as you type. The link in your address bar carries the scenario, so you can share exactly what you see.
Suppose robotaxis operate in the city.assumption
Each one drives km a day.assumption
A robotaxi uses about kWh per km.assumption
for one day. 100 km ≈ 1 home-day, rough benchmark.
About % of charging happens in a -hour daytime window.assumption
Sessions overlap, so peak is × the average.assumption
January adds % for cabin heat and cold batteries.assumption
Average daytime draw 117 MW, before winter.
A single urban feeder can realistically supply about MW before you are building a substation instead of connecting to one.assumption
A parkade depot holds about cars.assumption
Limited by power. 21 by grid capacity, 20 by parking.
Transmission-level interconnection. 5 to 7 year queue. Not feasible in a city core.
You cannot build one 164 MW site. You build 21 sites of 8 MW, and only certain pockets of the grid can take even that.
Fast chargers are kW each.assumption
× 150 kW. About 52 per depot.
Show the math
Energy scales smoothly. Buildable sites scale in steps.
Left axis: MWh per day. Right axis: depots needed at the current assumptions. The dashed line is your fleet. Every step on the right is a multi-year grid connection.
Every depot is its own grid connection.
Each depot draws 8 MW. That lands it in the 5 to 15 MW tier: dedicated feeder plus Connection Impact Assessment.
| Per-site load | What the utility has to build | Lead time, monthsassumption |
|---|---|---|
| ≤ 2 MW | Service upgrade off an existing feeder | to |
| 2 to 5 MW | Feeder upgrade | to |
| 5 to 15 MW | Dedicated feeder plus Connection Impact Assessment | to |
| > 15 MW | Substation bay | to + |
Tier lead times are Flock working assumptions pending P.Eng feeder review.5 Edit them if you know better.
Four other ways to serve the fleet. Each one runs into the same wall.
Every option below is priced against the scenario you set above. Change the fleet and these change with it.
One mega-depot outside the core
Put everything on cheap land km outside the service area. Each cab stops to charge times a day.assumption
8% of all fleet kilometres driven with nobody in the cab. The further outside the city, the more empty kilometres: every 10 km out is about 20 km of deadhead per stop.
And the site itself still needs 164 MW from one point on the grid. Transmission interconnection, 5 to 7 years, on top of the deadhead.
Use the public fast chargers
Send the cabs to the retail network everyone else uses.
Canada has 8,431 public DC fast ports in total.2 This fleet needs 13% of the whole country's supply, in one city, all day.
Retail ports serve one car at a time, with a driver present. Nobody cleans, inspects or resets the cab. Works for a pilot of dozens. Fails at hundreds.
Batteries on site to shave the peak
Size each grid connection to the daily average and let storage carry the daytime peak. Storage at $ per kWh installed.assumption
670 MWh of storage to bring 164 MW down to a 53 MW average connection.
Storage lowers the peak. It does not remove the connection, the site, or the years. A tool inside a depot, not a substitute for one.
The operator builds its own depots
Same physics, same city, same utility. The operator starts the search for land and feeder headroom when the permit lands.
Each one a 5 to 15 MW connection, 24 to 36 months, after site control and before construction. 36 to 60 months elapsed, per site.
This is the real competitor. It is not cheaper or faster; it just starts later. Whoever files first owns the headroom.
Secure the parkades and the feeder headroom now. Rent them to whichever operator arrives.
21 distributed sites inside the service area, each under the 8 MW feeder ceiling, cleaning and inspection on site, and the utility filings already in the queue. The operator's fleet lands on day one instead of year three.
The honest cost of going distributed: more sites means more of everything. Each depot carries its own cleaning bay, parts store, crew and floor space, so overhead scales with the 21, not with the fleet. Fewer, bigger sites would be cheaper to run. The feeder ceiling says you cannot have them, and the parkades already have the floor space.
The parkades already exist. This is what 21 of them cover.
Toronto Parking Authority runs 32 Green P garages holding 10,809 cars between them.6 The map picks the largest 21 as an illustration, garages first, then the biggest surface lots. These are not secured sites.
A cab returns to a depot within km of wherever it finished its last ride.assumption
Share of the city boundary within 3 km of a selected depot.
Against 10,000 cabs. Enough floor to stage the whole fleet.
Coverage clusters in the core, where the rides are. The gaps are where a depot search starts.
Candidate sites and capacities from Toronto Parking Authority open data (2019 snapshot). Coverage is a straight-line radius, not drive time. Map tiles © OpenStreetMap contributors.
Canada's ten largest cities hit the same wall. Here is how many depots each one needs.
Each of Canada's 10 largest metro areas gets the same robotaxis per resident as our Toronto scenario: 3.6 per 1,000 people. Every other assumption above carries over. This is a count, not a site search.
In the 10 largest metro areas. Every one a separate grid connection.
At 3.6 per 1,000 people, 20.6 million people.
Spread across 250 separate feeder connections.
Populations: Statistics Canada, 2021 Census, census metropolitan areas.7 The Toronto row is the whole metro, including Mississauga, Brampton and Markham, so it is larger than the city scenario above. Winter uplift is held at our Toronto figure everywhere; it runs milder in Vancouver and harsher on the Prairies. Smaller cities are left out: ride demand concentrates in dense metros, so scaling by population would overstate them.
Toronto has retail chargers. It has no fleet depots.
Regulation moves in years. Grid moves in more years.
Ontario's automated vehicle pilot runs through 2027. A permit in 2028 meets a three-year depot build. Whoever secures feeder headroom and parkade structure in 2026 is the only one ready. Flock is the landlord and pit crew for whichever operator arrives.
Talk to Flock.
We are assembling the depots before the fleets arrive. If that is a conversation you want to have, write to us. Deck available on request.
Start the conversationOr write directly: hello@driveflock.com
Where the fixed numbers come from
- Toronto vehicle-for-hire: about 1.1 billion km and 87 million trips in 2025, roughly one third of kilometres without a passenger, about 26,000 active vehicles. City of Toronto Vehicle-for-Hire report.
- Public DC fast-charging port count in Canada, Natural Resources Canada, March 2026.
- Toronto Hydro load capacity map and large-customer connection filing guidance.
- Waymo: roughly 263 km per vehicle per day; Miami depot about 12 to 18 months from land to operational.
- Tier lead times and per-site MW ceiling: Flock working assumptions pending P.Eng feeder review.
- Green P Parking, Toronto Parking Authority via City of Toronto Open Data, 2019 snapshot: 32 garages, 10,809 spaces. Coordinates and capacities as published.
- Statistics Canada, 2021 Census of Population: the 10 largest census metropolitan areas (20.6 million people) and the City of Toronto (2,794,356). Robotaxis per resident are taken from the Toronto scenario and applied to each metro.