[Ep. 087]

The Permit Is Part of the Product: Why Physical AI Wins or Loses at the Statehouse (Bradley Tusk)

Listen on:

The bottleneck is political, not technical

Start with energy.

Almost every Physical AI roadmap in energy and infrastructure rests on one assumption: the compute gets built. More data centers, more power draw, more grid. Bradley Tusk thinks that assumption breaks at the statehouse, not the server rack.

His argument is simple. Data centers push real costs onto local voters. Utility bills go up. And the politician who represents those voters notices.

"This is I don't want my voters to have to subsidize Jensen Huang and Sam Altman," Tusk said. "Why should my person who's an electrician pay 35% more in their utility bills just because these guys are doubling energy consumption?"

He is careful to separate this from NIMBYism. NIMBYism is about keeping people out. This is about who pays. When a hyperscaler's power draw shows up on a constituent's bill, the local official has every reason to say no to the permit and the zoning.

Tusk's read: "local politicians are going to realize this and say, I'm not giving you the permit for data centers."

He does not think that ends AI. He points to other paths. Inference-optimized compute instead of raw generative scale. Micro-grid nuclear. His point is narrower and sharper. The current build-out is planned around one form of compute and a mountain of debt, and it assumes permitting that may never show up.

For anyone building Physical AI in energy, grid, or infrastructure, that is the lesson. The hard part is not the technology. The hard part is the permit.

Regulation is a moat, not a cost

Most people treat regulation as a tax. Tusk treats it as territory.

He only invests in highly regulated companies. And he only takes equity when he believes his team can do two things: grow the company's TAM and build a regulatory moat against competitors.

"You have to believe and I have to believe that our work is going to meaningfully increase your TAM, build a regulatory moat for you against your competition," he said.

Flip the usual framing. If regulation is a compliance line item, it is a drag. If regulation is a wall you can build and your competitor cannot, it is a moat. The rulebook stops being defense. It becomes offense.

That reframing matters most in exactly the sectors where Physical AI lives. Energy. Defense. Public safety. Logistics. None of these are lightly governed markets. The company that learns to move the rules gets a lead that code alone cannot buy.

Actually changing the law, state by state

Here is where Tusk draws a hard line between what he does and what a big fund's regulatory operating partner does.

The mega-fund version, in his telling, is a glorified general counsel. They can help find the right lobbyists, the right PR firms, the right grassroots. Referrals, in other words.

His team does the thing itself. "What we do specifically is we go pass laws in states."

He gave one example. His team helped make Utah the first state to allow a prescription issued by AI, with no human in the loop. One portfolio company's AI can now identify a condition and send a prescription straight to the pharmacy. (The company name was garbled in the raw transcript, so treat that specific corporate detail as unverified. The mechanism is the point.)

Then he described the full toolkit. State by state by city, passes laws, blocks laws, writes regs, blocks regs, deals with procurement, deals with unions. That is a different job than knowing a good lobbyist. "I have yet to see a platform team at any fund that does that," he said.

For a founder deploying autonomous or physical systems, that distinction is the whole game. A referral gets you a meeting. A passed law gets you a market.

When to engage: the stakes gradient

Founders always ask Tusk when to start worrying about regulation. His answer is not a date. It is a question about consequences.

"Sometimes the answer is day one. Sometimes the answer is not until you have to."

What sets the timing is the penalty for getting it wrong. He put it bluntly. With scooters, the city impounds your hardware. Annoying, survivable. With regulated finance, you go to jail. Very different story.

So the framework is a gradient, not a rule. Map the downside first. If the worst case is a confiscated unit and a fine, you can move fast and engage later. If the worst case is criminal liability or a shutdown order, regulation is a day-one product decision, not a later cleanup.

For Physical AI, this maps cleanly. A delivery robot that gets ticketed on a sidewalk is one risk class. A drone in restricted airspace, an autonomous system on a public road, a device that touches patient care or critical infrastructure is another. The higher the stakes, the earlier the rulebook has to sit inside the roadmap.

His practical advice for founders who are not yet at the danger point: go meet your version of him before you need him. Build the relationship while there is nothing urgent to do.

Legacy industries are cracking open

Tusk ended on something more upbeat, and it lands right in dirty, dusty, dangerous territory.

He invests in legacy industries. Construction. Manufacturing. Transportation. Public safety. His framing: these are sectors that maybe missed the wave of SaaS but are now actually really opening up because of AI.

His example was voice AI for the trades. Now you can run a multi-million dollar HVAC business with a couple of employees, because all of your customer service is being done through voice. You still need people to fix the units. Everything around them gets automated.

Then he made a go-to-market point founders in these sectors underuse. Media is white space. In consumer, everyone fights for attention. In legacy industries, almost nobody is making content. He described one founder who sells to police agencies and started a podcast interviewing police chiefs and public safety leaders. Guests kept turning into customers. "You're not competing with 60 other podcasters," Tusk said. The media built credibility with the buyer before anyone tried to sell them anything.

The Dirty Jobs takeaway

For Physical AI founders in energy, defense, infrastructure, and public safety, the permit and the rulebook are part of the product.

You can win the demo and still lose the market if the zoning board, the statehouse, or the procurement office says no. In regulated industries, the regulation is not friction to route around. It is terrain to win. Sometimes the moat is the model. Often it is the law you helped write. Winning the statehouse can matter as much as winning the demo, so build both.

Pull quotes

"I don't want my voters to have to subsidize Jensen Huang and Sam Altman. Why should my electrician pay 35% more in their utility bills just because these guys are doubling energy consumption?"
"Local politicians are going to realize this and say, I'm not giving you the permit for data centers."
"Our work is going to meaningfully increase your TAM, build a regulatory moat for you against your competition."
"What we do specifically is we go pass laws in states."
"Sometimes the answer is day one. Sometimes the answer is not until you have to."

FAQ

Is regulation just a compliance cost for Physical AI startups?
Not in Tusk's view. He frames regulation as a moat and a TAM expander. If you can build a rule that helps you and blocks competitors, the rulebook becomes an advantage rather than a drag.

When should a founder start dealing with regulation?
It depends on the penalty for getting it wrong. Tusk's rule: sometimes day one, sometimes not until you have to. If the worst case is a confiscated device and a fine, you can wait. If the worst case is criminal liability or a shutdown, treat it as a day-one product decision.

Why does Tusk think data center build-out is a political risk, not just a technical one?
Because data centers raise local utility bills, and voters do not want to subsidize hyperscalers. He expects local politicians to block permits and zoning once bills climb. He points to inference-optimized compute and micro-grid nuclear as alternate paths.