Commodity

Some goods are the same everywhere. A kilo of wheat from France is, to a baker, the same as a kilo of wheat from Poland. Whether he buys in France or Poland therefore hardly depends on quality, but on price. Such goods are called commodities in English, or in German roughly “Massenware” or “Standardware.” The term comes from commodity trading, but today it’s used far beyond that. When journalists write that a product is “becoming a commodity,” they mean: it’s losing its distinctiveness and becoming interchangeable.

What interchangeability does to price

As long as a product is unique, the provider can charge a lot for it. After all, there’s no alternative. Once several providers deliver nearly the same thing, that changes. Customers then compare only prices and switch at the first cheaper offer. Profit per unit sold shrinks, often dramatically.

For companies this is a threat, for customers usually a blessing. Hard drive storage is a good example. In the 1990s, a gigabyte was worth a small fortune. Today storage costs so little that providers often throw it in for free. The technical progress was there, but there’s barely any money to be made from it anymore.

That’s why “commodity” is a warning word on the stock market. When analysts say a company’s main product is becoming a commodity, what they’re saying is: this company’s high profit margins are in danger. Stock prices often react sharply to such statements.

How an innovation turns into mass-market goods

It usually unfolds in four steps. At first, only a single company can do something, and it charges high prices. Then competitors copy the idea, because the knowledge spreads. After that, the products converge to the point where customers no longer notice a real difference. In the end, price remains the main argument left standing.

This is accelerated by open standards and by freely available software. When everyone is allowed to use the same blueprint, the barrier for new providers drops sharply. In AI, so-called open-weight models play a role here: AI systems whose fully trained innards anyone can download and run themselves. They drive down the price for all other providers as well.

Companies fight back by adding something that can’t be copied. That can be a brand, a particularly convenient access point, exclusive data, or a network of partners. A common misconception, by the way, is that a commodity is cheap to produce. Crude oil and computer chips are extremely costly to manufacture. Commodity doesn’t mean “simple” — it means “available everywhere, the same.”

Commodity in the AI business

The term currently comes up especially often with language models, i.e., AI programs that write texts and answer questions. A few years ago there were only a handful of providers for this, at high prices. By now, many models with similar performance exist, and the price per request has dropped sharply. Many observers therefore say that language models are becoming a commodity.

Not everything in the AI chain is affected by this. Specialized AI computing chips still come from almost only one manufacturer. As long as that remains the case, this part of the business is the opposite of a commodity, and prices stay high. That’s exactly why the stock market watches so closely whether competitors catch up.

You also encounter the word constantly outside of AI. Mobile phone plans, electricity, cloud storage, and flights are largely commodities: you mainly compare prices. Someone who buys headphones of a particular brand even though cheaper ones with the same technology exist, on the other hand, shows that this product is precisely not a commodity.