Flywheel Effect
The flywheel effect describes a self-reinforcing cycle: one success creates conditions that make the next success easier. In the tech industry, this usually refers to users who supply data that makes a product better — which in turn attracts new users.
A flywheel is a heavy disc that spins. At first you have to push it hard, and it barely moves. After many rotations it keeps going almost by itself, because the motion sustains itself. This exact image lies behind the flywheel effect in business. It refers to a cycle in which each step makes the next step easier. Success doesn’t just grow bigger — it also gets cheaper to achieve.
Why this makes leads hard to catch up with
Normally, a company grows for as long as it spends money on advertising and staff. A flywheel works differently. The lead a company already has generates the next lead itself. Whoever got there early has to invest less later than a new competitor does.
For investors, this is a key argument. They often pay high prices on the stock market for companies in which they perceive a running flywheel. The idea behind it: such an advantage doesn’t disappear easily just because a competitor builds a similar product. The product alone isn’t enough — the newcomer lacks the users and their data.
But the effect has an unpleasant flip side. A flywheel can also spin backwards. If users lose trust, the amount of data shrinks, the product gets worse, and even more users leave. This is then called a downward spiral, which is just as hard to stop as the upward motion was hard to slow down.
The cycle of users, data, and better models
For AI products, the cycle typically runs in four steps. A chatbot is good enough that people use it. These people leave traces: which answers they found helpful, where they had to ask follow-up questions, where they gave up. The company uses this feedback to retrain the program, i.e., improve it with new examples. The better product attracts even more users.
A second wheel turns alongside it when it comes to costs. Many users mean many requests, and that makes building your own data centers worthwhile. Large volumes lower the cost per request. Whoever can offer lower prices wins more customers and thereby becomes even cheaper.
It’s important to distinguish this from the network effect. There, the benefit to me increases because other people use the same platform — a messenger app is only useful with friends on it. With a flywheel, I don’t need to know the other users. They make the product better through the data collected, without my ever coming into contact with them. Both effects can occur simultaneously and add up.
Flywheels at Amazon, Tesla, and ChatGPT
The term became well known through management author Jim Collins and through Amazon. Amazon’s wheel works like this: more selection attracts more buyers, more buyers make the marketplace more attractive to merchants, more merchants increase the selection. Larger volumes simultaneously lower the cost per package, which allows lower prices.
In the AI industry, you mostly hear this word in quarterly reports and at developer conferences. Tesla uses it to justify its driver-assistance system: millions of cars sold provide video footage from real-world traffic. OpenAI points to the many ChatGPT conversations from which weaknesses in its own models can be identified. Chip developers like Nvidia tend to name their flywheel on the software side: whoever learns to program does so on the most widespread platform, and those programs then only run well there.
Still, be skeptical of this word. It sounds like a law of nature, but is often just a claim in marketing copy. A genuine flywheel is recognizable by the fact that growth costs fall while the product measurably improves. If neither happens, the wheel is only spinning in the presentation.