Analysts say AI data center spending needs much higher revenue

In short: Analysts are warning that the massive spending boom on AI data centers will only pay off if AI revenue rises sharply, and soon.

What's going on

Big technology companies are spending large sums on AI infrastructure, especially data centers (warehouses full of computers) and specialized chips. This wave of investment is so large that it is starting to affect debt markets, which is where companies borrow money.

Goldman Sachs estimates that the largest US “hyperscalers” (very large cloud computing companies like Amazon, Microsoft, and Google) could spend about $800 billion in capital spending in 2026, and around $1.1 trillion in 2027. Capital spending, often called capex, is money spent on long-term items like buildings and equipment.

Goldman’s analysts estimate these firms would need about $300 billion a year in AI-related revenue in the next few years just to break even, meaning to earn back what they spent. For comparison, Goldman says AI-related cloud revenue is currently running about $70 billion above the pre-AI trend. Goldman also estimates that to earn a 30 percent return on that invested money, data centers would need to bring in about $636 billion a year, which is far higher.

Other researchers put the needed revenue even higher. A paper by Stijn Van Nieuwerburgh estimates total AI data center costs could exceed $10 trillion from 2025 to 2032 and might require $3.7 trillion in annual revenue by 2032 to justify the spending under certain assumptions.

What to watch

A key question is whether people and businesses will pay enough for AI tools to cover these costs. If AI becomes cheaper to run, or if customers resist higher prices (like when a service that felt free suddenly gets expensive), the math behind today’s building boom could look less convincing.

Source: Financial Times