ChatGPT Ran the Numbers on Trump's Credit Card Cap

Trump wants a 10% interest rate cap on credit cards.

The industry charges 20%+.

This is going to hang over card issuers and banks for a while. What happens once we go lower than we have in decades?

Look, the goal of reducing credit card rates is a worthy one. But capping rates at 10% would inevitably cause millions of Americans to lose their cards. Credit card companies can’t profitably lend to subprime borrowers at that yield. The math just doesn’t work. You can’t make it up on volume. You especially can’t make it up on volume when the volume is people who don’t pay you back.

If this materializes (and the devil is in the details of any legislation) it would hit every issuer and bank. But what does that actually mean? “Bad for banks” is not a research opinion. A PM doesn’t want vibes. How do you quantify, quickly, who gets hit most and why?

That’s what I’ll show you.

I used a 2-step AI workflow to stress-test Capital One against a hard 10% cap. The framework works for any card issuer.

Step 1: What are the unit economics? Scrape filings, build a model.

Step 2: What does it mean? Translate into an investable view.

Let’s run through both.

Step 1: What Are the Unit Economics?

Before you can stress-test anything, you need to understand how the business makes money. For credit cards: revenue per dollar of loans, costs, losses, and what yield you need to break even.

This is exactly the kind of work that used to take a junior analyst a weekend. Now it takes a prompt.

I used ChatGPT 5.2 with thinking mode. It’s strong at structured modeling. Claude Cowork (which came out this week) also works, though you need to watch context limits.

The value is being able to drop in filings and have AI extract what matters. I fed it Capital One’s 10-Q, 8-K supplement, and earnings presentation. The prompt builds a mini P&L per $100 of receivables, calculates breakeven APR, maps credit quality, and flags what’s disclosed versus estimated.

Side note: if you don’t feel like scavenging the internet to find these documents, ask Gemini to fetch the links. Gemini has Google Search under the hood.

What came back was a structured Excel workbook. Seven sheets. Fully cited.

Here’s what the model shows:

Current Unit Economics

• Capital One earns a 18.0% yield on its card portfolio
• Funding costs eat 4.0%
• Operating expenses take 8.0%
• Credit losses consume 4.6%
• Breakeven APR: roughly 13.4%

At current yields, they make about $4.60 per $100 of loans. Profitable, but the cushion is thinner than you’d think.

The Sensitivity Is Brutal

• At 15%: Still profitable. About $1.61 per $100.
• At 10%: Very unprofitable. Losing $3.39 per $100.

That 10% scenario isn’t a haircut. Quarterly pre-tax swings from positive $3.1 billion to negative $2.3 billion. A $5.4 billion reversal. On one segment. In one quarter.

Credit Quality Mix

• 83% prime / 27% near-prime and subprime (below 660 credit score)
• This cohort gets cut if underwriting tightens

So now you have the numbers. But numbers don’t tell you what to do with them.

Step 2: What Does It Mean?

I use Claude for this because ChatGPT’s writing is horrendous. Claude, with the right context, is much better at reasoning and prose humans can digest.

The key is setting up a Claude Project. A regular chat won’t work because it doesn’t use RAG. RAG means retrieval-augmented generation. When you upload documents to a Project, Claude indexes and retrieves from them intelligently. In a regular chat, it stuffs everything at once and overloads the context limit (the model’s working memory).

I set up a Project with my analysis structure, dropped in the Excel plus source documents, and ran the synthesis prompt.

What came back was a full analyst write-up. The kind of thing you’d expect from a sector specialist, except that you don't have to be a financial institutions guru to analyze these opportunities because you have a smart little French guy named Claude by your side.

Here’s the summary:

The Verdict: SHORT. Policy beta: HIGH.

That’s not a recommendation. That’s what the numbers imply under a hard cap scenario.

Why Capital One Is Exposed

• 85% of Capital One’s revenue comes from interest income and they have a heavy revolver book. That means they’re super sensitive to interest rate caps. When your business model is “charge people 18% to borrow money,” a 10% cap is existential.
• Currently they yield 18% but breakeven from our analysis is 13.4%. A 10% cap doesn’t just hurt. It leaves them underwater.
• Credit cards are 76% of total company revenue. Not much opportunity to funnel customers to other products like personal loans (which in this case are already baked into the segment).

So what happens if this legislation actually passes?

• Management halts originations to high-yield revolvers
• Marketing gets cut, portfolio shrinks 10-20%
• Deep subprime loses access entirely
• Near-prime sees severe line cuts
• Prime is mostly fine

Where Do We Go From Here?

Capital One is a good test case because it has both cards and banking. You can run this framework on pure-play issuers and banks alike.

Synchrony is a private-label card specialist with heavier subprime exposure and an average 27% yield. A 10% cap would really hurt them. Bread Financial is similar.

But there could also be beneficiaries. SoFi’s personal loans run at 10-12% rates. Affirm’s BNPL products are installment loans, not revolving credit. Depending on how the regulation is defined, these names could gain as consumers get pushed out of traditional cards.

Run this on your list of names. Obviously check everything and apply your own diligence before placing a bet. But this gives you clear direction of where to dig.

And consider other prompts: what’s the probability legislation passes? What exemptions might exist? Which types of credit get affected? You can decompose all of this with AI.

That’s how you go from a headline where people blindly short every financial institution to one where you can quantify exposure and figure out who wins and who loses.

Personal Update:

I'll be in Asia for the next month meeting with our investment bank and hedge fund clients.

If you're around, hit reply!