Why Is Your AI Vendor Suddenly Selling Other Companies' Models?

At its developer conference last week, OpenAI launched a marketplace that lets enterprise customers spend their committed OpenAI dollars on third-party software and even on other labs' open-weight models, all purchased through OpenAI. On the surface it reads as generosity, the biggest lab in the world handing its customers the multi-model flexibility they've been asking for. Read it again. The commitment, the checkout, the routing, and the relationship all still run through one company, which makes this the smartest lock-in move of the year. We call the product rented optionality, flexibility you lease from the same vendor you wanted the flexibility from. It answers the board's lock-in question without reducing the lock-in by a dollar, and versions of it are about to show up in every renewal conversation you have.

What did OpenAI actually announce?

A marketplace where your spend commitment becomes a currency that works on other people's products.

The mechanics matter. Enterprises sign large spend commitments with AI vendors in exchange for discounts, and historically those committed dollars could only buy the vendor's own tokens. The new marketplace, launched with a few dozen partners, lets that same committed spend flow to partner software and to open-weight models from other makers, with OpenAI as the storefront. For the buyer, the appeal is immediate. One contract, one bill, one security review, and suddenly the committed money feels flexible. For the vendor, it's better than immediate. Dollars that might have leaked out to competitors now stay inside the ecosystem, the commitment gets easier to justify growing at every renewal, and each alternative you try through their store becomes one more reason the relationship is working. The genius of the move is that your experiments with other people's models now strengthen your primary vendor's account plan.

What is rented optionality?

It's the exit, leased from the party you'd be exiting.

We've written about rented control, the sovereignty features vendors lease to companies that want independence, priced as a premium tier of the very dependence they claim to reduce. Rented optionality is its sibling one layer up. Instead of leasing you control features, the vendor leases you the flexibility story itself, a menu of alternatives that lives inside their contract and dies with it. It's also distinct from the second hop, the middleman in your data path. This is a middleman in your purchasing path. You wanted a second source. You got a second aisle in the same store. The test for whether optionality is real is simple to state. Real optionality survives the relationship ending. Ask what happens to the marketplace purchases, the routing, and the committed dollars on the day you decide to leave, because that day is the only day optionality matters, and rented optionality evaporates on exactly that day.

Why will your own team love it?

Because it solves real, immediate problems, and all the costs are deferred.

The honest case for these marketplaces is strong, which is what makes them effective. Your team can taste new models without onboarding a new vendor, procurement gets one bill instead of six, the security review happened once, and finance already likes the discount attached to the commitment. Best of all, when the board asks whether the company is locked in, the team can answer that it runs models from several labs, which sounds exactly like the diversification everyone agreed to pursue. We've written about the seller's map, each vendor's drawing of the market where every road runs through their product. A marketplace is the seller's map made physical, the pitch about where the market is turned into the market itself, rebuilt inside the vendor's walls with your budget already in it.

What does it actually cost you?

The things optionality was supposed to buy, a credible exit, an independent router, private usage patterns, and leverage at renewal.

Walk through each. The exit erodes because the commitment grows, now that it covers more of your spend there's a case for making it bigger every year, and the share of your AI estate inside one contract climbs with it. The router is theirs, meaning your vendor now sees which alternatives you use, at what volume, for which workloads, which is precisely the intelligence a seller wants going into your renewal. Your purchases are subject to the storefront, and we've already watched access end over disputes between vendors that customers weren't party to, the pattern behind the availability clause. And the renewal leverage inverts, because walking away no longer means replacing one vendor's models, it means dismantling your whole purchasing arrangement, which is a bigger ask with a worse deadline. None of this requires bad faith from the vendor. The structure does the work by itself.

So should you refuse to use these marketplaces?

No. Use them as a tasting menu, never as a second source.

For experiments, short-lived workloads, and finding out which open-weight models actually fit your tasks, a marketplace is genuinely efficient, and refusing it on principle just slows your own learning. The mistake is letting it stand in for an optionality strategy, especially in front of a board that has asked the lock-in question and deserves a real answer. The working rule we give clients runs one way. Discovery happens through their marketplace. Dependency happens through your own seams. Any model that graduates from experiment to production earns a direct relationship, your own keys, or your own hosting, sitting behind routing you control.

How do you buy optionality that's actually yours?

Own the seam, hold one real second source, and put two questions to every flexibility pitch.

The seam is the architecture move we described in the model socket, a boundary you own that every model plugs into, so switching is a config change no vendor can see coming or price against. The second source is a direct relationship with at least one other provider, plus an open-weight fallback you could run if every storefront closed at once, kept warm with real traffic rather than listed in a slide. Then the two questions, for OpenAI's marketplace and for every imitation of it that follows. Who owns the router? What survives the day we leave? When both answers are the vendor, you're looking at rented optionality, and the price gets named at renewal, not at signing. Flexibility is now a product, and it's a good one. Just remember that optionality is the single thing you can't outsource to the party it's meant to protect you from.

If you want optionality you own, with the routing and the second sources mapped for your stack, start with an AI Blueprint or reach us at contact@theyor.com.