Note · Opinion, not rated
The rideshare map in 2031 is a manufacturing question.
Autonomous vehicles are about 0.9% of the US rideshare market today. Almost every forecast of where that goes argues about software. The binding constraint is how many vehicles can be built and how many cities will permit them, and both of those are countable.
Start with where things actually are
Waymo runs roughly 500,000 paid rides a week across ten US cities with a fleet of about 3,800 vehicles, and has raised $16 billion at a $126 billion valuation. Tesla's robotaxi is live in 7 metros, with its Texas fleet reported at 44 vehicles in mid-2026. Everyone else in the United States is at pilot scale.
The US rideshare market is roughly $60 billion of gross bookings a year. At about $20 a ride that is close to 58 million rides a week.
Where autonomy sits today
Waymo plus every other US autonomous operator, against the whole market
- Autonomous rides a week
- 0.52M
- US rideshare rides a week
- 57.7M
- Autonomous share
- 0.90%
- Rides per vehicle per week
- 132
That last number is the one to hold on to. It converts any share forecast into a fleet size, and a fleet size into a factory.
Why the growth curve is the wrong tool
Waymo went from 50,000 weekly rides in May 2024 to 500,000 in March 2026. That is roughly 216% a year, and it is genuinely remarkable.
Extend it five more years and you get 164 million rides a week, which is 234% of the entire US rideshare market.
A forecast that produces more than twice the available market has not told you about autonomy. It has told you that you used the wrong method.
This is the same failure as a discounted cash flow with a growth rate above the discount rate: the arithmetic runs away because nothing in it represents a limit. So instead of asking how fast the line goes up, ask what each destination physically requires.
What each 2031 share actually requires
| 2031 share | Rides/week | Fleet needed | vs today | Vehicle capex | Doubling time |
|---|---|---|---|---|---|
| 5% | 3.51M | 26,673 | 7× | $4B | 18 months |
| 10% | 7.02M | 53,346 | 14× | $8B | 12 months |
| 20% | 14.04M | 106,691 | 28× | $16B | 9 months |
| 35% | 24.57M | 186,709 | 49× | $28B | 7 months |
| 50% | 35.10M | 266,728 | 70× | $40B | 6 months |
Reading across the 20 percent row: it needs about 107,000 vehicles, twenty-eight times the current fleet, roughly $16 billion of vehicles alone, and the fleet doubling every nine months for five straight years.
The capital is the least interesting number there. Sixteen billion dollars is almost exactly what Waymo raised this year, and Alphabet could fund several times that without noticing. Money is not the constraint.
The constraint is the vehicles themselves and the operations around them. Each one needs a depot, charging, cleaning, maintenance, insurance, mapping, remote assistance and a local regulator who has said yes. None of that scales the way software does, which is precisely why Tesla has 44 vehicles in Texas and Waymo has 3,800 nationally despite Tesla having the louder story.
So the real question is how many get built
| Vehicles built per year | Fleet by 2031 | Rides/week | US share |
|---|---|---|---|
| 20,000 | 103,800 | 13.7M | 19.5% |
| 50,000 | 253,800 | 33.4M | 47.6% |
| 100,000 | 503,800 | 66.3M | 94.4% |
Twenty thousand purpose-built autonomous vehicles a year, sustained for five years, gets you to roughly 19.5% of the US market. Fifty thousand a year gets you to 47.6%.
For scale, twenty thousand units a year is a rounding error in automotive terms and a very large number in autonomous terms, because these are not ordinary cars. The sensor suite, the compute, the validation and the service infrastructure make each one closer to a small aircraft than a Camry.
Our actual view
We think national autonomous share of US rideshare in 2031 lands somewhere between 10% and 20%, and we would put more weight on the lower half of that range.
The reasoning is not sophisticated. Twenty percent requires a fleet doubling every nine months for five consecutive years with no regulatory setback, no safety incident that pauses a major market, and a manufacturing ramp nobody has yet demonstrated. Ten percent requires a doubling every twelve months, which is close to the current rate and therefore the one we would bet on.
But the national number is the least useful way to think about it, and this is the part most coverage gets wrong.
The map will be extremely uneven
Autonomy does not roll out nationally. It rolls out city by city, and within a city, neighborhood by neighborhood. The economics favor dense, warm, grid-planned metros with cooperative regulators, which is exactly why the first ten cities look the way they do.
So the honest 2031 forecast is not a single share number. It is a bimodal map:
In perhaps fifteen to twenty-five US metros, autonomous share of rideshare will be very high. Plausibly a third to a half, possibly more in the dense cores where the service area is fully covered. These will be the places where a human driver becomes the premium option rather than the default.
In most of the country it will be approximately zero. Rural areas, small cities, anywhere with severe weather, anywhere the per-vehicle economics cannot support a depot. The fixed costs of launching a market do not fall just because the software improved.
A national figure of fifteen percent is the average of those two worlds, and it describes neither.
What this means for the incumbents
Uber and Lyft are not obviously the losers. The received view is that autonomy destroys them, and it might. But demand aggregation and autonomy are different businesses, and the current partnerships, Waymo with Uber in Austin and Atlanta, May Mobility with Lyft, suggest the fleet operators would rather rent demand than build it.
The question is who captures the margin. A network that supplies riders to a fleet it does not own earns a take rate. A fleet owner who must also acquire riders carries the customer acquisition cost. Whichever side is scarcer in 2031 gets the economics, and today vehicles are scarcer than demand.
The driver is the clearest loser, and that deserves stating plainly rather than being hidden inside a share forecast. In the metros where this works, driving for a living stops being viable, and it stops over a few years rather than a generation.
What would make us wrong
In the direction of faster. A genuinely cheap purpose-built vehicle, produced on an existing automotive line rather than retrofitted, would break the manufacturing constraint and the table above with it. If a credible operator announces a hundred thousand units a year, our range is too low.
In the direction of slower. A serious safety incident in a major market. Any outcome that forces per-city relitigating of permissions already granted. Or utilization falling as service areas expand into thinner demand, which would raise the fleet required for any given share.
And in a direction we have not modelled at all. Our entire frame assumes autonomy competes for the existing rideshare market. If driverless cost structures make the service much cheaper, the market itself grows and takes trips from personal car ownership and transit. In that world the share question matters far less than the market-size question, and every number in this note is too small.
That last possibility is the one we would want to be wrong about, and the one a reader should push on hardest. We have forecast a share of a market we assumed stays roughly the same size, which is the most conventional assumption in this note and quite possibly its weakest.