Game of Tariffs

AI Simulates the 2025 Tariff War—And Reveals What Might Happen Next

Have you seen the latest geopolitical thriller? It's got betrayal, retaliation, and one very angry supply chain.

We’re now nearly two weeks past Liberation Day, when the Trump administration launched a sweeping tariff campaign on nearly all U.S. imports—10% across the board, plus targeted retaliatory threats against China, the EU, and others. We are still unsure about where the numbers came from, but as I mentioned last week, AI may have been involved.

What is clear is this:
Traditional economic logic is not what’s driving this.

Instead, we’re seeing the use of tariffs as tools of coercion, loyalty testing, and signaling. It's not surprise that analysts and politicians alike are increasingly taking this perspective. Senator Chris Murphy recently described as “political weapon.” Others, like the game theorist Sebastian Moritz, have argued this is a multi-level loyalty game: pressure foreign governments and business leaders into falling in line—or paying the price.

For decision-makers across policy and business, or anyone trying to understand the strategic direction of U.S. policy: this isn’t just theory—it’s a map of how this conflict could spiral, recalibrate, or stall.

Important caveat: the fact that something looks like a strategic game doesn’t mean the person playing it is a master strategist. All politics can be modeled as game theory—not because everyone is a genius, but because everyone is reacting to incentives, power, perception, and incomplete information.

In other words, just because you're playing chess, it doesn't mean you're Garry Kasparov. A lot of people play chess. Not all are good.

That said, I still want to know more. So I sought Dr. Gemini and Dr. ChatGPT's help and asked:
What games are being played? And what are the likely outcomes?

And a question I didn't ask AI but am putting up front for all of us to ponder: given that AI may have been involved in the tariff numbers, do we think AI could be involved in the administration's strategy? Again, we may never know, but it would be fascinating, and if the administration did do so, what AI came up with to answer this question could give us a hint as to what they would get if they did, in fact, use it for strategic purposes.

Gaming the Tariff War

I'm not a game theorist or an economist. But I am a policy person with access to some powerful AI tools and a lot of curiosity.

So I teamed up with Gemini Deep Research and ChatGPT-4o, prompting them to model the post-tariff landscape using the tools of game theory—a field that studies how people (or nations) make decisions when outcomes depend on how others behave.

I fed them public economic timelines (notably from PIIE and trade press), and asked them to simulate the strategic behavior of six key actors:

• China

• The European Union

• Canada and Mexico

• U.S. domestic markets

• U.S. political leadership

Each relationship was mapped onto a classic game-theory structure:

• Repeated Prisoner’s Dilemma for China (retaliation and trust)

• Signaling games for the EU (will a pause be read as weakness?)

• Coordination games for Canada and Mexico (retaliate or preserve USMCA?)

• Principal–Agent games for Trump and his voter base (short-term wins vs. long-term cost)

• Market signaling games for Wall Street (reacting to ambiguity and tweets)

Then, I asked the models to simulate what’s most likely to happen—strategically, economically, and politically—based on the incentives and constraints in each arena.

Trump offers ‘buy’ tip on his social media platform, “Truth Social” hours before tariff pause that made stocks soar. This is an example of what “Market signaling” entails.

What the AI Predicted

Each simulation represents a different “game,” with its own dynamics, players, and likely outcomes. Here are some high-level insights with some of my initial reactions:

• U.S.–China: A repeated retaliation game. If trust breaks once, it may never recover.

  • Comment: Appears tit-for-tat on paper, but exemptions show there’s a performance layer to the escalation. A game is being played—but not necessarily well.

• U.S.–EU: A signaling game. Will the 90-day pause be interpreted as diplomacy—or capitulation?

  • Comment: A pause may look conciliatory, but if the U.S. reads it as weakness, it’s a classic example of why signal interpretation—not intent—drives missteps.

• U.S.–Canada & Mexico: A coalition game. Can allies coordinate without fracturing USMCA?

• U.S. Domestic: A principal-agent game. Trump's political incentives may diverge from national economic interest.

  • Comment: The logic here looks strategic, but keep in mind: playing the crowd doesn’t mean you’re winning the game.

• Markets: A real-time signaling game. Tweets move markets more than fundamentals.

  • Comment: Markets don’t wait for 4D chess—they react to noise. This is one arena where the perception of chaos is its own risk.

Below are two tables:

The first summarizes what AI thinks is most likely to happen in each arena, including strategic forecasts, policy risks, and recommended moves.

Note: Just because a dynamic fits a game-theory model doesn’t mean it’s being played with strategic depth. Real-world actors often bluff, improvise, or misread the board.

The second table highlights likely sectoral impacts—from consumer prices to tariff revenue—across industries like autos, electronics, and agriculture.

The downstream effects of these games are felt at the cash register—even when the moves are more symbolic than substantive. Some outcomes may stem less from economic design, and more from theatrical miscalculation.

My Two Cents (Free Version)

After running these simulations through Gemini and ChatGPT, my first reaction was this:
The output is impressive—but also incomplete.

Gemini synthesized over 100 sources and mapped out structured game-theory scenarios across five arenas. The result reads like a strong graduate-level policy memo. But as with most AI-generated analysis, what’s most interesting is what it didn’t do.

For one: the models treated each arena—U.S.–China, U.S.–Domestic politics, U.S.–Markets—as siloed. In reality, these games are tightly interdependent. Domestic political incentives shape China strategy. Market reactions reshape diplomatic signaling. None of this happens in isolation.

Second: the models missed the nuance of strategic exemptions, like the U.S. electronics carve-outs that softened the apparent China escalation. That move fundamentally alters the game—but doesn’t show up in the “paper” version of the conflict. Looking at the sector predictions table shows where exemptions blur the line between strategy and spin. The model sees impact; the politics may delay it.

Still, if you’re clear-eyed about the limitations, this kind of modeling remains a valuable tool—especially when trying to anticipate strategic misreads, retaliatory loops, or sector-level shocks.

What Trump is doing right now—being publicly tough while quietly carving out exemptions—is a classic two-level game. It lets him score political points while softening the economic blow. That’s strategic—but it’s also standard. Electronics imports make up a massive share of consumer goods exposure—so exempting them radically changes inflation dynamics and consumer pain, and thus limiting potential domestic backlash.

Want more of my thoughts on how this logic plays out in real time—and how complexity doesn't equal brilliance?

Want my full commentary on where the models fall short, what they still get right, and how to use this kind of AI wisely?

Subscribe below for the full breakdown.

Behind the Paywall…

Subscribe for full access to the 70+ page AI-generated game theory simulation and my strategic breakdown of what it got right—and where it missed the mark.

Behind the paywall, you'll get:

My Commentary

• Which game models held up—and where they collapsed.

• Where AI overfit the logic, ignored coalition dynamics, or failed to integrate simultaneous games.

• How tariff exemptions (like those in electronics) challenge the core predictions.

• What the simulations got right—and how to read them without getting misled.

Strategic Deep Dives

• Game-by-game analysis across five arenas: China, the EU, Canada & Mexico, U.S. domestic politics, and financial markets.

• What kind of game each player is in (retaliation, coordination, signaling, principal-agent), and what that implies for policy stability and global risk.

• Side-by-side comparisons of three future scenarios: Escalation Loop, Strategic Pause, and Multilateral Recalibration.

Sector-by-Sector Impact Forecasts

• Import volume loss and CPI inflation by sector (autos, electronics, apparel, agriculture, steel).

• Tariff revenue projections under different elasticity assumptions.

• Long-run substitution effects and pass-through risks.

Behavioral and Institutional Risk Mapping

• Where diplomatic signals (like the EU’s 90-day pause) can backfire.

• How market reactions, media narratives, and election pressures shape game outcomes.

• Structural weaknesses in trade agreements exposed by short-termism.

AI Capability Assessment

• How AI simulated the logic of trade wars—and what it struggled to capture (e.g. emotional retaliation, reputational risk).

• Commentary on belief-updating, trust erosion, and the epistemological limits of model-based policymaking.

Policy Playbooks

Tailored takeaways for:

• U.S. and allied trade negotiators

• Multilateral coalition builders

• Market actors facing signal ambiguity

• Firms navigating price shocks and supply chain risk

Includes a downloadable PDF of the full research brief created by Gemini Deep Research, ChatGPT-4o, and Claude 3.7 Sonnet, with strategic framing and editorial oversight by me.

This isn’t just a model. It’s a map of what to watch next.

Subscribe to get it.