Picture a Goldman Sachs conference room sometime before the launch. Someone from Apple is explaining that the card must be titanium. Not because titanium makes a better payment instrument — a magstripe is a magstripe — but because when you drop it on a restaurant table, it needs to make a specific sound. Someone from Goldman, an institution that had spent a century and a half not caring what its products sounded like on furniture, has to decide whether that is a reasonable request.
They said yes. Fifteen years later, that decision has an ending: in 2026, Apple Card moved from Goldman Sachs to JPMorgan Chase, with a 24-month transition period. Goldman’s consumer banking adventure closed a chapter, as CNBC put it in January.
I review AI tools for a living, which means I spend most of my week watching companies make the exact mistake Goldman made. So let me be clear about which mistake I mean, because it is probably not the one you think.
The numbers were not the problem
Apple Card had 3.1 million users by March 2020. By early 2024, 12 million. That is roughly a 4x increase in under four years for a credit card in a saturated US market. Almost no consumer financial product grows like that. If you handed a growth chart like that to any product team without a logo attached, they would frame it.
And the partnership still ended. That gap — great top-line growth, partnership dissolution — is the whole story, and it is the story I keep watching replay in the AI tooling space with different names on the jerseys.
Because growth was never the thing Goldman was buying. Goldman was buying a position: consumer bank, at scale, fast. What it actually got was the role of balance sheet and regulatory shell for someone else’s product, where the someone else owned the interface, the brand, the customer relationship, and the definition of what “good” meant. Apple decided the card was titanium. Apple decided there were no fees. Apple decided the app looked like that. Goldman decided how to eat the consequences.
Why this is an AI tools story
Half the AI companies I evaluate are in the Goldman seat and do not know it.
The pattern looks like this. A startup builds real infrastructure — inference, orchestration, retrieval, evaluation, something with actual engineering behind it. Then a much larger platform offers distribution. Suddenly the startup’s product is a component inside someone else’s experience. The larger platform owns the surface the user touches. The startup owns the part that breaks at 3 a.m.
The growth chart looks incredible. Usage goes up and to the right in a way no amount of independent marketing could achieve. And every quarter, the startup has slightly less say over its own roadmap, because the roadmap is now a downstream function of a partner’s product decisions.
Here is the part that should make founders uncomfortable. Goldman Sachs is one of the most sophisticated counterparties on earth. It has lawyers whose lawyers have lawyers. It knew how to read a contract. It still spent years in a partnership where the other side held the customer relationship, and then it exited. If Goldman could not negotiate its way out of that structural position, a Series A company with 40 employees and a term sheet it really wants to sign is not going to do better.
What I look for now
When I evaluate an AI product that has a marquee distribution partner, I have started asking a different set of questions:
- Who owns the interface the end user actually looks at
- Who decides pricing, and who absorbs the cost of that pricing decision
- If the partnership vanished tomorrow, does this company have customers or just usage
- Whose name is on the support ticket when something fails
- Can the partner replace this company without the user noticing
That last one is the killer. When the Apple Card issuer changed, reporting noted existing users would not see visible change. Think about what that means. The entire banking relationship — the underwriting, the balance sheet, the regulatory apparatus — swapped out underneath 12 million people, and the plan was for it to be invisible. That is what being a component looks like from the outside. Replaceable by design, and the replaceability is marketed as a feature.
The uncomfortable read
I do not think Apple did anything wrong here. Apple did exactly what a company with an enormously valuable customer relationship should do: it rented the boring, capital-intensive, heavily regulated part from whoever offered the best terms, kept the part customers love, and swapped vendors when the terms stopped working. That is competent business.
The lesson is for everyone else. Distribution from a giant is not a gift, it is a trade, and the thing you trade is usually the customer relationship. Sometimes that trade is correct — plenty of companies should take the volume and the revenue and be happy. But you should know you made the trade.
Fifteen years, two issuers, one card. The titanium outlasted the partnership. It was always going to.
🕒 Published: