What if the most important part of Waymo’s self-driving stack isn’t the software at all?
Because the news this week isn’t about a new model, a better sensor fusion pipeline, or a lower disengagement rate. It’s about money spent on people in Washington. Waymo, the Alphabet-owned autonomous vehicle developer, has doubled its lobbying spending as it pushes U.S. regulators to clear a path for fully autonomous taxi services. It’s calling for a federal framework to ease the rollout of robotaxis. And it’s doing this in direct competition with Uber.
I spend most of my time here evaluating AI tools on things I can actually test: latency, accuracy, cost per call, how badly the docs lie to you. Lobbying disclosures are not on that list. But I’d argue they should be, at least for this category, and that’s a genuinely uncomfortable thing for a reviewer to admit.
The part of the product you can’t benchmark
Here’s what makes robotaxis different from every other AI product I look at. If I review a coding assistant and it’s bad, you don’t use it. If I review a vector database and it’s slow, you pick another one. The market sorts it out in weeks. Autonomous taxis don’t work that way. A robotaxi that performs beautifully in a city where it isn’t legally allowed to operate is worth exactly zero to a rider.
So the addressable market for this technology isn’t defined by capability. It’s defined by permission. Which means the spending on federal lobbying isn’t a side activity next to the engineering work. It’s the thing that determines whether the engineering work gets to exist commercially.
The scale of this is worth sitting with. According to Streetsblog Empire State, Uber and Waymo together spent more than $15 million lobbying New York politicians this year. That’s one state. The federal doubling is on top of a state-level effort already running into eight figures for a single jurisdiction.
What a federal framework actually buys
Waymo’s stated ask is a federal framework, and I want to be fair about why that’s a reasonable position rather than just a cynical one. Right now, an autonomous vehicle operator negotiating city by city and state by state faces a patchwork of rules, each with its own approval timeline and its own definition of what counts as safe. That’s expensive, slow, and it punishes whoever is furthest along in actual deployment.
A single national standard would replace that patchwork with one set of requirements. If you’re a company with a working product and a compliance team, that’s clearly better. If you’re a city that wants specific control over what drives on your streets, it’s clearly worse. Both of those things are true simultaneously, and the lobbying money is the mechanism for deciding which view wins.
Reading this as a tool evaluator
I keep coming back to what this means for how anyone should assess AI companies operating in regulated territory. A few things I’d hold onto:
- Regulatory position is a moat, and it’s a durable one. A competitor can match your model in a year. Matching your relationships with policymakers and your accumulated approvals takes considerably longer.
- Spending on compliance signals seriousness, not weakness. Doubling a lobbying budget is a company telling you it expects the regulatory question to be the deciding one. That’s usually an accurate read of its own situation.
- Two companies fighting over rules will produce rules that suit those two companies. Waymo and Uber are the ones at the table. Whatever framework emerges will be shaped by their operational realities, not those of a startup with three cars and no lobbyist.
- Capability benchmarks tell you less than you think here. Ride availability in your city is a better signal of a robotaxi product’s maturity than any published safety metric, because availability already includes the regulatory approval.
The honest verdict
I don’t think there’s a villain in this story. Waymo wants clearer rules so it can ship faster. Uber wants rules that suit its own model. Both are behaving exactly as you’d expect companies with billions invested and a fragmented approval process to behave. The broader industry push toward established operational frameworks for autonomous vehicles is, in some form, necessary. Somebody has to write these rules.
What I’d push back on is the framing that treats this as boring corporate news adjacent to the real technology story. It is the technology story. The reason you can hail a robotaxi in one city and not another has more to do with lobbying disclosures than with model weights.
For anyone tracking this category, add the disclosure filings to your reading list alongside the technical blog posts. They’re less fun. They’re also more predictive.
đź•’ Published: