AI data centres spending on chips and memory surge

Five charts to start your day

For $10 a month, or $100 a year, you support a simple mission: spread great data visualisation wherever it comes from. You help fund the work of finding, sourcing and explaining the charts that deserve a wider audience. And you back a publication built on generosity, transparency and the belief that better understanding makes a better world.

CHART 1 • AI data centres spending on chips and memory surge

AI still gets discussed through model launches and benchmark scores. But once a model is released, the question changes: how much capacity can the company actually supply, and at what price?

This chart estimates memory rising from 2% of data-centre build cost in 2021 to 18% in 2026. Processors remain the largest share, but memory is no longer a minor line item.

That is why Anthropic’s Fable 5 rollout is a useful caveat, not a clean proof. The company says demand is high and capacity is hard to predict, while users see a powerful model being capped, rationed and moved towards usage credits. Maybe that reflects scarce compute; maybe it is also pricing power. Either way, the best model still has to be served through expensive infrastructure.

Source: Mike Zaccardi

I like this chart because it makes AI less mystical. Behind every answer from a model sits a chain of suppliers, data centres, power contracts and capital budgets.

Paid subscribers get access to the other four charts: record power-sector M&A, the US-China split in top AI models, the world’s most valuable unicorns and Amazon’s AWS-and-advertising profit engine. Together, they show why the AI race is shifting from demos to infrastructure, capital and distribution.