Monoculture

The term originates from agriculture. A field on which only a single plant species grows is called a monoculture. Such fields are easy to farm, but a single pest can destroy the entire harvest. In technology, the word describes the same pattern: a great many people or companies use exactly the same software, the same devices, or the same provider. If this one system fails or has a flaw, everyone is affected at the same time. Diversity, by contrast, works like insurance: if one solution fails, the others keep working.

When a single bug brings half a country to a standstill

Monocultures don’t arise from foolishness but from rational choices. A single system is cheaper to purchase, easier to maintain, and all employees already know it. That’s exactly why the market leader’s market share often keeps growing. Eventually an entire industry ends up depending on one product, without anyone ever having planned it that way.

The problem only becomes apparent when things go wrong. In July 2024, the security company CrowdStrike distributed a faulty update. Millions of Windows computers subsequently failed to start. Airports, hospitals, and banks came to a standstill worldwide – not because the bug was particularly sophisticated, but because so many used the same software.

With AI, a second effect comes into play. When authorities, companies, and schools all use the same language model, they also inherit its blind spots. A model that systematically rates certain applicants worse will repeat this error identically everywhere. With ten different models, such biases would at least partially cancel each other out.

Why everything keeps converging on a handful of providers

Technology markets tend toward concentration on their own. Once a provider has many users, it earns more, invests more, and thereby becomes even more attractive. Experts call this a network effect. On top of that comes the effort of switching: anyone who has stored data in one system for years finds it hard to get back out.

With AI, costs reinforce this pull. Training a large language model costs hundreds of millions and requires data centers full of specialized chips. Worldwide, only a handful of corporations can manage that. Even providers who seemingly offer their own product often rely on someone else’s model in the background.

This creates monoculture on several levels at once. At the very bottom sit a few chip manufacturers, above them a few cloud providers – that is, companies that rent out computing power. Above that lie a few large models, and thousands of apps are built on top of those. An outage right at the bottom can reach all the way to the top. This is an important difference from an ordinary market leader: in a monoculture, everyone shares not just the provider but also its vulnerabilities.

Monoculture in the news, in politics, and on your own phone

In business news, the word usually appears in connection with outages and antitrust proceedings. When a cloud service fails and streaming services, online shops, and banking apps stop working at the same time, that’s a monoculture effect. Discussions about Nvidia also show this pattern: one manufacturer supplies the vast majority of AI chips worldwide.

Politically, the antidote is often called digital sovereignty. This refers to the attempt not to be completely dependent on providers from a single country. That’s why the EU promotes open models and its own data centers. Open models are ones whose building blocks are publicly available and that anyone is allowed to operate themselves.

In everyday life, you encounter this topic more subtly. Many school essays, job applications, and news articles are now created with the help of the same two or three chatbots. As a result, phrasing and arguments increasingly resemble each other. A common misconception is that monoculture is simply another word for monopoly. A monopoly describes one provider’s market power, whereas monoculture describes the lack of diversity and the resulting vulnerability.