How the AI Industry Is Organised

The layers of the AI industry, from chips and cloud computing to models and applications, and where profits and power concentrate.

Artificial intelligence, especially the generative AI that can write text, create images and write code, has become one of the fastest-growing industries in the world. Economists often describe it as a stack of connected layers.

The layers

Chips: specialised processors, especially graphics processing units, or GPUs, that perform the huge numbers of calculations AI requires.
Cloud computing and data centres: massive facilities full of servers where AI models are trained and run. Amazon, Microsoft and Google dominate cloud computing.
Foundation models: large AI models trained on vast amounts of data, which can be adapted to many tasks. They are developed by companies such as OpenAI, Google, Anthropic, Meta and others, as well as Chinese firms like DeepSeek.
Applications: products built on top of models, such as chat assistants, coding tools, customer service systems and tools for doctors or lawyers.

Where the money flows

In the early years of the generative AI boom, much of the profit went to the lower layers, especially chip makers like Nvidia and cloud providers, because everyone building AI needed their products. Model developers spent enormous sums on computing and talent, often running large losses while growing rapidly.

Economics of scale

Training a frontier AI model can cost hundreds of millions of dollars or more, mainly for computing power. Once trained, a model can serve millions of users, spreading that fixed cost. This creates strong economies of scale, favouring large, well-funded firms.

Rapid change

The industry changes fast. Costs of using AI models have fallen sharply as competition and efficiency improved. In early 2025, the Chinese company DeepSeek released a model reported to match leading models at much lower training cost, briefly shaking investor confidence in chip and AI companies.

Building on the stack

A small Indian start-up wants to offer an AI tutor in Hindi. It does not build its own chips or data centre, and it may not train its own large model. Instead, it rents cloud computing, uses an existing foundation model and builds a specialised application. Each layer of the stack lets smaller firms build on the work of larger ones, while paying fees along the way.

Thinking AI is a single product

AI is an industry with many layers, each with different economics. Chip makers, cloud providers, model developers and application builders face very different costs, competition and profits.

Key takeaways

The AI industry includes chips, cloud computing, foundation models and applications.
Early profits concentrated in chip makers and cloud providers.
Training frontier models is very expensive, creating economies of scale.
Costs of using AI have fallen quickly as competition and efficiency improved.