Control · Fall 2026Page 86 of 100.THE REVIEW
A long, clean corridor lined with tall black server cabinets, with a monitor cart at the far end.
A long aisle of server cabinets in a data center. Hao’s argument is that products which feel weightless on a screen depend on buildings, land, power and water.The photograph is by Brett Sayles on Pexels.

Review

Hao follows the supply chain, and the magic trick falls apart.

Karen Hao’s Empire of AI is a reported history of OpenAI that argues the industry behaves like an empire, and it makes the case more convincingly than its occasional overreach would suggest.

Written by CONTROL

This article runs 477 words and takes about two minutes to read.

8out of 10
Book
Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI
By
Karen Hao
From
Penguin Press, 2025

Most books about artificial intelligence are written from inside the demo. Karen Hao’s is written from the far end of the cable. Empire of AI, published by Penguin Press on May 20, 2025, is a history of OpenAI built on roughly 260 interviews, about ninety of them with people who worked at or near the company, and it won the National Book Critics Circle Award for nonfiction in March 2026. Sam Altman declined to take part.

Hao first profiled OpenAI for MIT Technology Review in 2020, when it was still presenting itself as a nonprofit research lab, and the company was not pleased with the result. The book picks up that thread and follows it through the rise of ChatGPT, the departure of Elon Musk, the brief firing and quick reinstatement of Altman in November 2023, and the growing unease of researchers such as Ilya Sutskever about where the company was headed. The boardroom chapters are brisk and well sourced, and they will be the reason many readers pick it up.

Rows of network cables plug into a rack of labeled switch modules.
Network cables bundled into a rack of switches. Empire of AI follows the cable out of the demo and toward the people and places it quietly depends on.The photograph is by Sergei Starostin on Pexels.
A woman holding a tablet leans against a glass wall beside rows of server racks in a corridor.
A woman with a tablet leans against the glass wall of a server room. The boardroom chapters draw readers in, but Hao’s subject is the human machinery behind the text box.The photograph is by Christina Morillo on Pexels.

The argument from the edges

They are not the reason to keep it. Hao’s real claim is structural. She argues that the leading AI companies behave like empires: they take resources they do not own, including the text and images of the internet; they rely on cheap labor in poorer countries; they concentrate expertise; and they justify all of it with the rhetoric of a race that must be won. To make the case she leaves San Francisco for Kenya, where contract workers were paid to label graphic and disturbing content so that chatbots would not repeat it, and for Latin America, where communities have fought data centers over water.

Those chapters change the way the product looks. A reader who has only ever met these systems as a text box on a screen comes away with a sense of the physical and human machinery behind it, and of how little say the people supplying it have had.

The empire frame can be too tidy, and at times Hao treats disagreement about the technology’s promise as bad faith. The book also contained a real error. In December 2025 she corrected a passage about a proposed Google data center near Santiago, Chile, after a reader found that a government document she relied on had mislabeled its units, which had overstated the facility’s water use relative to the local population by a factor of a thousand. The corrected figure still shows the center using slightly more water in a year than the district’s eighty-eight thousand residents, but the episode is a reminder that a polemic needs its numbers exact.

The verdict: It is a clear, well-reported account of how the current AI boom was built and who pays for it. Read it with a skeptic’s pencil, which is how a book this pointed asks to be read.