A small group of companies controls key parts of the AI supply chain

In short: The tools needed to build and run advanced AI are increasingly controlled by a small number of companies, across chips, cloud computing, and the biggest AI models.

What's going on

AI does not come from one supplier. It comes from a chain of parts and services, like how cars depend on steel, factories, and dealerships. In AI, the key layers include chips, cloud data centers (warehouse sized computer rooms), and the large AI models that many apps use.

At the chipmaking end, a few firms act as bottlenecks. ASML, a Dutch company, is effectively the only maker of EUV lithography machines (the ultra precise tools used to make the most advanced chips). TSMC in Taiwan manufactures around 90% of the world’s most advanced chips, and NVIDIA is estimated to control about 80 to 90% of advanced AI chips used in data centers.

On the cloud side, five companies, Amazon, Google, Meta, Microsoft, and Oracle, are estimated to own about 71% of the world’s cumulative AI computing capacity as of late 2025, measured in comparable high end NVIDIA chips. Many major AI labs rely on these companies to rent computing power, which can affect who can afford to train and run the largest models.

At the model layer, a small set of companies, including OpenAI, Anthropic, Google, Meta, Microsoft, and Amazon, provide foundation models (general purpose AI systems that other apps plug into, like an engine you can reuse).

What to watch

More governments are paying attention to this concentration, because it can shape what information people see through chatbots, search, and workplace tools. Watch for new rules, and for efforts to build alternative chip supply and computing capacity outside the current centers of control.

Source: NYTimes