Every hotelier knows the feeling. It’s the second week of a soft month, the pickup report looks thin, and someone in the meeting says it out loud: “The hotel across the road just dropped their rate.”

The room goes quiet. Everyone’s doing the same math. If we hold, we might lose share. If we follow, we protect occupancy tonight, but we all end up selling the same rooms for less. And the next month, we’ll be having this exact conversation again, just from a lower starting point.

So when I read the AI headlines this month, I didn’t see a tech story. I saw a revenue meeting, played for much higher stakes.

The same dilemma, at trillion-dollar scale

On September 12, Anthropic CEO Dario Amodei published an essay proposing a slowdown in AI development among the labs building the world’s most capable models. Within a day, OpenAI’s Sam Altman and xAI’s Elon Musk had publicly agreed.

That sounds like good news. And maybe it is. But the obvious question is the one we ask in every revenue meeting: what if the other side doesn’t hold?

Companies will be reluctant to limit the performance of their models in the name of safety for fear of losing ground to competitors, and officials have also rejected calls for a slowdown, citing fear of losing ground to other countries. 

Replace “AI lab” with “hotel” and “other countries” with “the new opening down the street,” and you’ve got a conversation I’ve had in Bangkok.

Game theory 101: the Prisoner’s Dilemma

Economists have a name for this. It’s called the Prisoner’s Dilemma, and it goes like this: two players can each cooperate or defect. If both cooperate, both do well. If both defect, both do badly. But if one cooperates while the other defects, the defector wins big and the cooperator gets crushed.

The trap is that, individually, defecting always looks smarter. Whatever your rival does, you’re better off cutting. So both cut. And both end up worse off than if they’d simply held.

In AI, “defecting” means racing ahead while the other side pauses. In our world, it’s dropping rate while the comp set holds. There’s even academic studies on the AI version, which show that the more competitors there are, and the less they trust each other, the more corners get cut.

Sound familiar? A market with two luxury hotels behaves very differently from one with twelve.

But maybe it’s not a Prisoner’s Dilemma at all

Here’s where it gets interesting, and where I think most commentary misses the point.

There’s a second game called the Stag Hunt. Two hunters can team up to catch a stag, a big reward, but only if both commit. Or each can go off alone and catch a rabbit, a small but guaranteed meal. Nobody wants the rabbit. They settle for it because they’re afraid the other hunter won’t show up.

The difference matters enormously. In a Prisoner’s Dilemma, you need to force cooperation which of course we cannot and should not do; there are also antitrust laws which stop competitors from agreeing on prices. In a Stag Hunt, you only need to build trust. The players already want the better outcome; they just need assurance.

And I’d argue most hotel rate wars are Stag Hunts. No GM in Bangkok actually wants to discount. We all know it trains guests to wait for deals and erodes what they believe the experience is worth. (This is exactly what my doctoral research on perceived value keeps showing: once a guest’s reference price drops, it’s very hard to walk it back up.) We cut because we’re scared of being the one left holding rate while everyone else fills up.

How you fix a Stag Hunt

Look at what the AI slowdown proposal actually asks for. It centers on a three-step plan: embedding independent third-party safety reviewers, establishing coordinated industry safety standards within democratic nations, and eventually securing global agreements, including strict limits on AI chip exports to companies and countries that do not agree to prioritize AI safety.

Read that through a game theory lens and it’s a textbook answer:

Monitoring. Outside reviewers mean everyone can see whether the others are keeping their word.

Assurance. Shared standards tell each player what “holding” actually means.

Consequences. Export limits make defection costly.

Good hotel markets run on the same three ingredients. We monitor through rate shopping and market reports. We build assurance by being consistent, so the market learns we hold; they’ll develop confidence or event trust that you will not drop rates and so hopefully they will not either. And the consequence of chasing the bottom is built in: a hotel that discounts its way to occupancy usually wakes up to a weaker brand, lower guest expectations and a team that feels it.

Talk is cheap. Signals aren’t.

Of course, anyone can say they’ll slow down. The skeptics have been quick to point out the timing. OpenAI’s IPO has been pushed to 2027, while Anthropic is still expected to begin marketing its IPO in mid-October at the earliest and complete the listing before the November midterms. Critics ask, reasonably, whether the calendar is doing some of the talking.

Game theorists separate cheap talk from costly signals. Words cost nothing. Actions that hurt a little are what make others believe you. Anthropic paused parts of its AI training and testing after its models took unauthorized actions online, and moved about 150 engineers onto security, which is the kind of move that carries more weight than any essay. Whether that’s enough, time will tell.

In hotels, the costly signal is holding rate through a soft weekend and living with the occupancy dip. Your competitors notice. Once. Twice. By the third time, they start believing you.

The part nobody likes to say out loud

There’s a twist in the AI story that every hotelier will recognize immediately. If AI companies coordinate a slowdown, they might be breaking US antitrust law, because restricting market output, even for safety reasons, can look a lot like collusion to a regulator.

Welcome to our world. We can’t sit down with the GM across the road and agree to hold rate. Nor should we. The same goes for talent: AI labs poach each other’s researchers the way we poach each other’s chefs and sales leaders, and no-poach agreements are off the table too.

So cooperation has to happen without a handshake. That means it has to come from somewhere else: discipline, consistency and a genuine belief that we’re competing on value, not price.

What this means for us

The lesson I take from all this is simple. The best way to win a rate war is to change the game you’re playing.

If you treat your market as a one-off Prisoner’s Dilemma, you’ll cut every time and so will everyone else. If you treat it as a repeated Stag Hunt, where you’ll face the same neighbours for the next twenty years, you start playing differently. You invest in the product instead of the discount. You make your value visible. You hold, and you hold again, until holding becomes what the market expects of you.

The AI labs are trying to figure out whether they can trust each other enough to slow down. We’re trying to figure out whether we can trust each other enough to hold rate. Different industries, same question.

So here’s mine for you: in your market, are you playing a Prisoner’s Dilemma or a Stag Hunt, and what would it take to change the game?