Kenyan data workers and Chilean water activists share this audiobook with Silicon Valley engineers, which is the argument compressed into a casting decision. Karen Hao began covering OpenAI in 2019 believing the nonprofit framing, and the book is the record of that belief coming apart. What replaced it is a thesis about scale: the vision requires compute, data, cheap human labor to clean that data, and quantities of energy and water large enough to become somebody else's problem. Empire is not a metaphor she reaches for at the end. It is the organizing frame from the first chapter.
Hao's Access Reached the Altman Firing
Access is the asset that separates this from the commentary shelf. Hao was reporting on the company before it became the story, kept her sourcing intact through the years that followed, and was therefore close to the sudden firing and reinstatement of Sam Altman when it happened rather than reconstructing it afterward from other people's coverage. That closeness is what the book spends its authority on, and it is the reason the corporate-governance chapters read differently from anything assembled at a distance.
Gossip is how an August 2025 four-star opens, and it is a fair warning wrapped in a joke. That response credits it with a closer look at OpenAI's internal workings than any comparable book in that reader's experience, along with real domain knowledge and years of reporting for major outlets, then says plainly that the book has an agenda and strays from impartiality. Both halves are correct. Anybody buying this expecting a neutral corporate history has misread the title.
Press quotation fills the opening of the published description before a word about the contents arrives: a magazine's hundred-name list, a critic at a national paper, a culture site, a prize, two best-of-year mentions. It is an unusual amount of decoration for a book with this much reporting inside it, and buyers should scroll past it. The material underneath is considerably more interesting than the wall of endorsement standing in front of it.
Five Costs Replace Boardroom Personalities Here
Compute, data, labor, energy and water are the five costs the book keeps returning to, and structuring an argument around inputs rather than personalities is what makes it durable. Anyone can write about a boardroom. Following the electricity to a substation and the water to a community that has to live with the draw, then setting that beside the wage paid to somebody labeling harmful content, produces a picture that survives the next round of executive musical chairs.
January 2026 brought a five-star that reads the book the right way, crediting its perspective and the nerve behind the reporting while noting the bias in plain terms and deciding to hear it anyway. That is a mature response to committed journalism. The book does not pretend to neutrality, the reporting is verifiable, and a listener who can hold both facts at once will get far more from the book than one who needs the author to have no position.
Seventeen hours and fifty minutes makes a long shift for any reader, and this one is delivered by the reporter who did the work. Self-narration suits a book whose authority rests on personal access, since the person saying she thought they were the good guys in 2019 is the same person who watched that assumption fail. Journalists are not trained voice actors, and across a runtime this size the trade is intimacy against polish.
This Response Gets Cut Off Mid-Thought
One five-star on this page is not really a response at all. Filed in February 2026, it arrives as a structured abstract, running through consolidation of power, extraction of data and labor, and a comparison to historical colonialism, then breaking into a heading for key themes and literary devices before the text is cut off. Whatever produced it, a prospective buyer learns nothing about the experience of listening from that entry.
Arithmetic on this page fails at the first step. The counter records exactly one rating and the display reads 4.4, which no single vote can produce, because one vote is one whole number of stars and nothing between them. Five people have written about the book, so the pool and the counter are visibly out of step as well. Treat the score as decoration and read the written material instead.
Length and repetition are the fair criticisms of the book itself. The empire frame is powerful and it is applied consistently, which by the second half means arriving at conclusions the argument has already taught you to reach. Chapters set in Kenya and Chile carry the most new information. Chapters inside the company carry the most momentum. The stretches that connect the two occasionally restate a case already made.
Supply-Chain Reporting Goes Deeper Than Usual
Technology reporters, policy people and anybody working adjacent to model development will find this the most useful long-form reporting on the subject so far, and in particular the supply-chain material that rarely gets pursued to this depth. Readers of investigative business books will recognize the form. Listeners who came for an explanation of how large language models work should understand that this is a book about the company and the extraction, not about the technology.
March 2026 produced the one response that touches the listening experience, a three-star saying the material was good to know and difficult to listen to. That verdict does not separate subject from delivery, and both readings are plausible: this is distressing content at considerable length, read by its author. Anyone weighing the purchase should sample the recording rather than assume the complaint is about one or the other.
Approach with care if you want a balanced account with equal weight given to the industry's own case, since the book states its position and argues it. Somebody who already agrees with the thesis will still find new reporting, which is the stronger case for spending the time. Anyone who follows this industry through company blog posts and conference keynotes should hear what the same story sounds like from the substation and the labeling floor.
Frequently Asked Questions
Does Karen Hao narrate the Empire of AI audiobook herself?
She does, and it fits a book whose spine is her own reporting history with the company from 2019 onward. A five-star filed on March 3, 2026 calls the result both informative and enjoyable, which is the harder combination to hit at this length. Expect a journalist's delivery rather than a performer's, sustained across nearly eighteen hours.
Is the audiobook technical, and do you need machine learning background?
No background is required. The subject is corporate power, resource extraction and the human supply chain behind model training, not architecture or mathematics. Terms like compute and large language model are explained in passing. Anyone with no technical training whatsoever will follow every chapter without difficulty.
Does the audiobook try to be impartial about OpenAI?
No, and it does not claim to. The framing of empire, extraction and colonial comparison is stated openly in the description and carried through the reporting. What you get is a documented case rather than a balanced survey, so the useful question is whether the evidence holds up, not whether the author is neutral.
