Vendor lock-in

Vendor lock-in is the situation in which a buyer keeps using one company's database, chip or software platform, even after a better or cheaper one arrives, because the buyer's own code and data are built around it. A vendor is the company that sells the product. The cost of moving from one vendor's product to another is called the switching cost. When the switching cost is larger than what the buyer would save by moving, the buyer stays. That is lock-in.

Here is how it builds up. A company chooses a product and starts to use it. Its programmers write code that uses the product's own features, such as special commands or settings that other products do not have. Its data is saved in the product's own format. Its staff learn the product, and its other systems are connected to it. Each year, there is more code, more data and more connections. To move, the company would have to rewrite the code and copy all the data to the new product without losing or changing any of it. It would have to test everything again and retrain its staff. It would also have to run the old and the new products side by side for a while, in case something goes wrong. For a large company, this can take years. So the first choice of a database or a platform is often the hardest one to reverse.

Lock-in is not always a mistake. A product's own features are often the reason it was chosen, and using them well saves time. The risk comes later. The vendor knows that its customers cannot leave easily, so it can raise its prices. Buyers protect themselves in a few ways. They choose products that follow open standards, so that another product can read the same data. They keep their data in open formats, such as the open table formats used for tables in a data lake. And they choose open-source software, whose code anyone can download, so that more than one company can support it.

Take, for example, a large retailer with fifteen years of sales records in one data warehouse, and thousands of reports built on it. A newer warehouse may cost less each year, but rewriting and checking every report can cost more than the saving.

Consider an AI team whose models are written with one GPU maker's tools, the GPU software stack. A rival chip is useful to that team only if its maker also builds an AI compiler that runs the same models without rewriting them.

Say a company runs hundreds of virtual servers on one vendor's hypervisor, and the vendor's new owner raises the price sharply. Many companies in exactly this position recently began to look for another hypervisor for the first time in twenty years.

How is vendor lock-in different from a network effect? A network effect makes a product more useful as more people use it, as with a payments app that more shops accept. It is a reason to join. Lock-in is about the cost of leaving: the code, data and skills that one buyer has built around one product. A product can have both. Lock-in is also one reason for an installed base, the systems that stay in use for decades once they are in place.