AI ecosystem · Hyperscalers · Neoclouds · AI labs · Colocation · as of —

Datacenter

Who is building AI data centers, with which CPUs, GPUs and custom chips, how many gigawatts they have and plan, what they spend, where their revenue comes from, and what analysts expect for 2026, 2027 and 2028.

Hyperscaler revenue: this year and next

Total company revenue expected for the current fiscal year and the next, with growth. Microsoft, Oracle and Alibaba use their own fiscal years (labelled); the others are calendar years.

Consensus estimates, except Alibaba's next year, which extends its consensus at the ~12% a year growth analysts expect.

Power: online now vs. target

Gigawatts each company has running (solid) against the target it has stated or that analysts model (dashed). Hover or tap a row for detail, click to open the company card.

Capex this year

Capital spending guided for 2026 (or the fiscal year covering most of it), at the midpoint, and the contract backlog that spending is meant to serve, with the clients behind it.

Contract backlog and the clients behind it

Company by company

The data-center-linked revenue line is highlighted in blue in each revenue bar.

What the chips cost

Chip makers do not publish list prices. These are analyst estimates and street prices, and cloud rental rates for comparison.

Published on-demand or committed prices, Sep 2026. Rubin, GB300, Trainium3 and MI455X have no public hourly price yet.

Sources