$110bn. That’s Waymo’s valuation after finalising a $16bn funding round, according to the Financial Times. Uber, meanwhile, has pledged $10bn to win the robotaxi race. And Waymo has doubled its lobbying spend while reportedly exploring a split from Uber entirely.
I build bots for a living. Not cars, but the same species of problem: an autonomous system that has to sense something, decide something, and act on it without a human babysitting each step. So when I read numbers like these, my first instinct isn’t to think about sensors or neural nets. It’s to think about who controls the interface between the bot and the customer.
Two companies, two very different bots
Strip away the branding and you have two systems that were, until recently, complementary parts of one pipeline.
Uber’s system is a matching and dispatch engine. Given a rider, a location, and a pool of available vehicles, produce an assignment. It has spent over a decade tuning that loop against real demand, and it holds the app that riders already have installed.
Waymo’s system is a driving stack. Given a road, produce safe motion. That’s a genuinely harder engineering problem in the narrow technical sense, and Waymo has poured years into it.
For a while, plugging one into the other looked obvious. Waymo drives, Uber fills the seats. The reported tension, and Waymo’s exploration of a split, suggests at least one party decided that arrangement wasn’t worth what it cost.
Why the split makes sense to a bot builder
Anyone who has shipped a bot that depends on someone else’s platform knows this feeling. You build the smart part. Someone else owns the door the user walks through. And every quarter, that someone else gets a little more comfortable reminding you whose door it is.
I’ve watched this play out at much smaller scale. You write a solid automation that runs beautifully inside a third-party messaging platform, then the platform changes its rate limits, or its pricing, or launches a first-party feature that does 70% of what your bot does. Your engineering quality never mattered as much as your position in the stack.
Waymo doubling its lobbying spend is the same instinct expressed in a different currency. Regulation is the other interface layer nobody owns outright yet. Who can operate where, under what permits, with what reporting requirements, and whether a company that builds the driver can also run the fleet. Those rules get written once and then shape the economics for a decade. Spending money to influence them is arguably cheaper per unit of outcome than shaving another few percent off your perception error rate.
The architecture lesson buried in the headlines
Here is what I take from this as someone who designs systems rather than corporate strategy.
Own your ingress
The hardest technical component is not automatically the most valuable one. Value tends to pool where the user actually arrives. If your bot’s only path to users runs through a platform that could plausibly compete with you, that’s an architectural risk, not just a business one. Build an abstraction layer over that channel early, even if you only ever use one implementation of it. The cost of the interface is low. The cost of being locked in is not.
Vertical integration is a bet on control, not efficiency
Running your own dispatch, your own app, your own support, and your own driving stack is more work than partnering. The payoff isn’t efficiency. It’s that no single external decision can strand you. That’s the tradeoff Uber’s $10bn commitment and Waymo’s possible exit both appear to be making, from opposite directions. Uber wants the driver it doesn’t have. Waymo wants the demand channel it doesn’t own.
Capital is a feature
$16bn buys iteration cycles. In bot development, the equivalent is having enough runway or compute budget to run experiments that fail. Small teams tend to optimise for never being wrong, because being wrong is expensive. Well-funded teams optimise for finding out fast. That difference compounds, and it’s usually a bigger predictor of outcome than raw talent.
What I’d actually watch
The interesting question isn’t which company wins. It’s whether the driving stack becomes a commodity that gets licensed out, or stays a proprietary moat tied to a single fleet. That distinction decides whether independent builders ever get an API to work against, or whether autonomous transport stays a walled garden with two gates.
The reported facts point toward walls going up rather than doors opening. A doubled lobbying budget, a $10bn counter-commitment, and a partnership under strain aren’t the signals of an ecosystem preparing to open up.
For those of us building smaller bots, the transferable insight is unglamorous but useful. Spend real effort on the boring layers: your integration boundaries, your fallback paths, your ability to swap out any dependency you don’t control. The clever part of your system is rarely what determines whether it survives.
đź•’ Published: