The biggest technology companies now report AI infrastructure as one of their largest expenses. Understanding what that money buys explains much of today’s market news.

Where the money goes

  1. Accelerators. GPUs and custom AI chips are the single largest cost in a modern AI cluster.
  2. Networking. High-speed interconnects tie thousands of chips together so they can train one model.
  3. Buildings and cooling. Dense AI racks generate far more heat than traditional servers, pushing operators toward liquid cooling.
  4. Power. Large campuses need grid connections, substations and sometimes on-site generation.

Why spending keeps rising

  • Training runs grow. Each frontier generation has typically used more compute than the previous one.
  • Inference grows with usage. Every chatbot answer and every AI feature in an app consumes compute. As adoption spreads, inference becomes a permanent, growing cost.
  • Reasoning models that “think” longer per answer use more compute per request.

The energy question

The International Energy Agency’s 2025 report Energy and AI projects that electricity consumption by data centers worldwide will roughly double by 2030, with AI as the main driver. The impact is concentrated: a few regions host most large data centers, which strains local grids and permitting.

Why markets care

Infrastructure spending creates winners along the supply chain — chip designers, memory makers, power equipment suppliers and utilities — but it also raises a question investors ask every quarter: will revenue from AI products grow fast enough to justify the investment? That debate drives much of the volatility in AI-related stocks.

What to watch

Capital expenditure guidance from the largest cloud providers, utility deals for new capacity, and efficiency improvements that could reduce the compute needed per task.

What a large AI data center contains

  • Accelerators (GPUs or custom chips) — the most expensive component.
  • High-speed networking to connect thousands of chips so they can train models together.
  • Storage for datasets and model checkpoints.
  • Power infrastructure: substations, transformers, backup generators and batteries.
  • Cooling systems, increasingly liquid cooling for dense AI racks.
  • The building and land, often in locations chosen for power availability.

Chips can be replaced every few years as new generations arrive, while buildings and power infrastructure last decades — an important difference for how investments are valued.

How it is financed

Large technology companies fund most of their spending from operating cash flow, but the scale has led to more use of debt, partnerships with infrastructure investors, and leasing arrangements with specialized data center developers. Some AI companies rent capacity from “neoclouds” — cloud providers specializing in GPU computing.

The bull and bear cases

Optimists argue AI demand will keep growing as models become more capable and widely used, and that infrastructure built now will be fully utilized. Skeptics worry that spending is running far ahead of AI revenue, that efficiency gains could reduce the need for compute, and that some investments may not earn adequate returns — comparing the situation to past infrastructure booms such as the late-1990s fiber-optic buildout.

How to read a capex announcement

When a company announces a spending figure, check whether it covers one year or several, whether it includes all data centers or only AI, and how much is chips versus buildings and power. Compare it with the company’s revenue and cash flow, and with what it said in previous quarters — changes in guidance often move markets more than the headline number itself.

Read more about how much energy AI uses, liquid cooling in AI data centers and why NVIDIA dominates AI chips to understand the physical side of the AI boom.

Key takeaways

  • Capital expenditure on AI data centers has risen dramatically among large technology companies.
  • Spending covers chips, networking, power, cooling and buildings.
  • Chips depreciate quickly; buildings and power infrastructure last decades.
  • Energy availability increasingly determines where data centers are built.
  • Markets debate whether AI revenue will justify the investment.

Frequently asked questions

What does capex mean?

Capital expenditure: spending on long-lasting assets such as buildings, equipment and infrastructure.

Why do investors care about AI capex?

Because it affects profits, cash flow and the returns companies will earn if AI demand grows as expected.

Could an AI investment bubble happen?

Some analysts warn of overinvestment; others see sustained demand. History shows infrastructure booms can include both lasting value and losses.