Rent the AI Engine, Own the Car: How to Avoid the Vendor Lock-In Trap

Every company I talk to is falling in love with a supplier that already has a plan to raise their rent. They just haven't gotten to the part of the plan that involves you yet.

The romance runs the same way every time. The tokens are cheap, sometimes free. The onboarding is smooth. The support engineer answers on a Saturday. You build a few workflows, then a few dozen, and pretty soon the whole company runs on a vendor you were kicking the tires on last spring. Everybody gets comfortable. And comfort, in this business, is the sound of a door clicking shut behind you.

Missed Part 1? We recently looked at the macro crunch hitting the AI industry in The VC Party Is Ending: What Happens When AI Tool Pricing Actually Has to Turn a Profit?. But if rising hardware costs and energy crises are the macroeconomic threat, there is an even bigger operational trap snapping shut on your business right now.

We've seen this movie enough times to know how it ends. Microsoft gave Windows away to anyone with a beige box and a floppy drive, and a generation of businesses woke up one day unable to imagine doing payroll on anything else. Google handed you Gmail and a terabyte of storage for the price of reading your mail, and now your entire professional identity lives on a system you couldn't leave in under a month if your life depended on it. Amazon rented you a server for pennies, then built a toll booth so your data checks in and never checks out.

The pattern is older than most of the people currently getting rich from it: Give the thing away, wait for everyone to settle in, then start turning the screws.

The Subsidy Is Not a Gift

Here's the part people skip. Nobody hands you a frontier model for the cost of a sandwich because they're generous. As we established when looking at the upside-down margins of major providers, these labs are torching tens of billions a year. That money comes from investors who would eventually like to be paid back.

The cheap token you're enjoying today is a marketing expense. It is the sample tray at the grocery store, and the sample tray is not the business model.

The business model arrives later, once you've rebuilt your operations around one company's quirks, one company's formats, and one company's little proprietary conveniences that felt so helpful right up until they became the bars of the cage.

That's when the price sheet changes. Not with a bang, just a brief email about “updated pricing to better reflect the value we deliver.” And you'll pay it, because by then your alternative is ripping out six months of work and explaining to your board why the thing that was supposed to save money is now eating it. Your supplier eats your margins, and the funny part is you handed them the fork yourself.

What Actually Belongs to You

So here's the thing worth getting straight before the screws turn: The model is not your asset. The model is a rental car.

What belongs to you, the actual assets you should be guarding like they are the only things in the building worth stealing, is everything around the model. It is the prompts you've tuned, the handoff documents that turn a vague ask into a clean output, the templates, the physical documents, and the workflows your people have learned to drive. Those are portable. Those don't care which engine is under the hood.

Think about what a good handoff prompt actually is. It's the accumulated judgment of someone who's done the job a hundred times, written down so precisely that any competent model can pick it up and produce the right thing. That document took real work and real expertise to build, and it will still be worth exactly what it's worth to you whether the engine behind it is American, Chinese, or something that doesn't exist yet.

Nobody can raise the price on a thing you already own. That's the whole game, and almost nobody's playing it.

Don't Fire Your Humans

Which brings me to the advice nobody selling you a platform wants you to hear: Don't fire your humans.

The move everyone's being sold, where you gut the team and let the model run the shop, is the exact move that welds you to a single vendor forever. Because once the humans who knew how to prompt, probe, and shape the output are gone, you lose the ability to switch. You're not running the tool anymore; the tool is running you, and the meter is spinning.

Keep the people. Have them build the artifacts and the handoff prompts so that the intelligence lives in your process, not in somebody else's cloud.

Do that, and the day a competitor shows up doing ninety percent of the job at a tenth of the price (thank you, papa Xi Jinping) with none of the proprietary hooks sunk into your side, you swap it in over a weekend. Your eyes won't water when you open the invoice. Your artifacts and your handoff prompts are important. Your preferred model is not.

This Ends as a Utility

Zoom out far enough and you can see where the whole industry is headed, because it's headed where every general-purpose network technology has ever headed. The railroads got trust-busted. Electrification got regulated into a boring, reliable commodity you don't think about until the bill comes. Ma Bell got broken into pieces because a single company owning the way a whole country talks to itself turned out to be a bad idea.

Intelligence delivered over a wire is going to walk the same road. It's too fundamental to leave in four or five hands, and the people who run those hands know it, which is why they talk so much about safety and so little about pricing.

The tell is the barrier to entry, and for niche work, it's already on the floor. You don't need a frontier model to draft your contracts, clean your data, or answer the same forty customer questions you get every day. You need something competent and cheap, and there's more of that every month. So as the big general models jack up prices to pay the investors back, the natural gravity of the thing pulls everyone toward the efficient supplier down the street. That's not a prediction. That's just water running downhill.

And if you think the code and the weights are going to stay locked in a vault, notice that they already haven't. Anthropic had its Open Claw moment. OpenAI had DeepSeek.

Both times some noble-minded insider decided the world's leverage shouldn't sit inside a single building, and out it went. I'm not going to pretend I'm scandalized. There's a strong argument that the most dangerous concentration in this whole story isn't the technology, it's the ownership, and every leak chips a little off the monopoly before it hardens.

Rent the Engine, Own the Car

Which is the mindset I'd leave you with. Treat the model like electricity. You want it clean, you want it cheap, you want a second provider you can flip to when the first one gets greedy, and you absolutely do not want to build your house in a way that only works if one particular power plant stays online and stays friendly.

The companies that come out of this era with their margins intact won't be the ones who bet biggest on a single lab. They'll be the ones who kept their people, built their own artifacts, and stayed light enough on their feet to walk when the price got stupid.

Falling in love with your supplier is a fine way to feel modern for about eighteen months. Then the rent comes due, and love was never in the contract.

Disclaimer/Full Disclosure (You made it!): This blog post was generated with the assistance of AI, with N&L human oversight ensuring accuracy and insight. The thoughts and opinions expressed are our own.