provider dependency

in English: vendor lock-in

basicwidespread

A situation where your work relies so heavily on one provider's product that switching would be too expensive, slow, or risky.

In plain terms

Vendor lock-in is like when only one manufacturer's capsule fits your coffee machine. The machine was cheap, but from then on, you do not decide at what price and from whom you buy the coffee. If the capsule price rises, you either pay or throw away the machine.

In the AI world, the same happens, only instead of a capsule, there is a language model, a cloud service, or an agent platform. A company builds its processes on it, trains its staff, and uploads its data. Later, it realizes that exiting would mean a months-long, expensive transition.

This is not necessarily a malicious trap. Often, it is precisely the most convenient, best-integrated solution that binds you the most. The cost of dependency becomes apparent when the provider raises prices, discontinues a feature, changes the terms, or when a better competitor appears.

How it works

Easy entry: providers lure users with free trials, discounts, or simple onboarding, prompting quick adoption.
Integration: more and more tools are being built to fit the provider's own solutions. These include instructions tailored to the model (prompts), agent configurations, connections to other systems, and functions that work only in that specific environment.
Accumulation: the stored data, conversation history, and colleagues' habits increase. These are difficult or impossible to transfer one-to-one elsewhere.
Switching cost: if you want to leave, you have to rewrite, test and retrain everything. Data export fees and penalties under long contracts may add to this. Under the EU Data Act, cloud providers may not charge any fee for switching from January 12, 2027. Until then, they may charge at most their actual costs.
Hold-up: since the provider knows you would find it hard to leave, they can dictate prices and terms. You can at most adapt.

An everyday example

An accounting firm entrusts invoice processing to the agents of a single AI provider and, over two years, adapts all its templates and rules to them. When the provider doubles the subscription price, it turns out that switching to another system would take months of work. The firm therefore chooses to pay the higher fee.

Who is affected

entrepreneur, executiveYou bear the risk if your company's operations depend on the pricing and decisions of a single AI provider.

developerYour design decisions determine whether the system can later be easily switched to another model or platform. → Code

employeeIf your knowledge is limited to operating a single tool, part of your routine becomes obsolete when switching providers. → Work & HR

everyoneAs a private individual, you encounter it when your photos, notes, or AI conversations get stuck in a single company's system.

A common misconception

MythIf I choose an open-source or standard solution, I am not threatened by provider lock-in.

In realityAn open solution reduces risk but does not eliminate it. The knowledge required for operation, custom developments built on it, and habit can bind you just as much.

Limits and risks

Avoiding dependency entirely is rarely worthwhile, because a system designed for multiple providers is more expensive and more complex. The goal is rather a deliberate, acceptable degree of dependency.
Language models cannot be swapped one-to-one. The same prompt may yield different quality results with another model, so everything must be retested when switching.
Data exportability alone is insufficient if another system cannot meaningfully read the format.
The provider can unilaterally change prices or terms, or discontinue a model, so it is advisable to prepare an exit plan in advance.

Similar, but not the same

monopoly – In a monopoly, there is no other provider in the market; in vendor lock-in, there is, but switching to them is too expensive for you.
commitment period – A loyalty period is a contractual obligation that expires, whereas provider dependence is technical and organizational, and typically strengthens over time.
network effect – In network effects, you stay because everyone else is there; in vendor lock-in, you stay because your own investment would be lost.

Related concepts

Broader

switching cost, piaci verseny, digital sovereignty

Narrower

cloud dependency, model dependency, data lock-in

On the edge

interoperability, data portability, open source, open standard, multi-vendor strategy

A brief history

The phenomenon was already known in computing during the mainframe era. Economics has dealt with switching costs since the 1980s. Its business significance was widely popularized by Carl Shapiro and Hal Varian's 1998 book "Information Rules." The European Union's Data Act, adopted in 2023, already facilitates switching between cloud providers by law.

Where things stand

At the end of 2026, provider dependency is a frequently cited business risk of AI deployments, especially for systems built on models and agent platforms. The cloud-switching rules of the EU Data Act have applied since September 2025. Open interoperability standards are also spreading, such as MCP, which connects agents with external tools. In practice, however, switching remains costly.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

OpenAI and four rivals publish Agent Plugins, a common extension formatAug 10, 2026 · Code

Also known as

vendor lock-in · lock-in · vendor lock-in · vendor dependence · provider lock-in · vendor dependency · proprietary lock-in

Sources

Wikipedia: Vendor lock-in

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.