Parmy Olson spent years covering technology at Bloomberg and Forbes before writing this, and it shows in the most important way: the book is reported rather than synthesised. There are people in it who talked to her, documents she obtained, and detail that did not come from a press release. That distinguishes it sharply from the large and growing category of AI books assembled from public coverage and rearranged into a narrative. The structure is a double biography.
Sam Altman at OpenAI and Demis Hassabis at DeepMind, two organisations founded with explicit commitments to developing artificial general intelligence safely and for humanity's benefit, both of which ended up inside enormous advertising-funded corporations. OpenAI's entanglement with Microsoft, DeepMind's acquisition by Google. The book is about how that happened, and it is a better question than the one most AI books ask. The contrast between the two men does a lot of structural work without being reduced to a simple opposition.
Hassabis comes through as the researcher who wanted to solve intelligence and then use it to solve everything else, genuinely academic in temperament, repeatedly frustrated by the commercial gravity of his parent company. Altman comes through as the operator, brilliant at capital and coalitions and at moving before anybody else has decided, with a relationship to the safety commitments that the book treats sceptically and does not caricature. The supporting cast, Musk, Thiel, Zuckerberg, Brin, is handled with restraint, appearing where they matter rather than because they generate attention. The central insight, and the reason this book will still be worth reading in five years, is structural.
Both organisations made real commitments about how this technology should be developed. Both then needed compute at a scale only a handful of companies could supply. The commitments did not survive the dependency, not because anybody dramatically abandoned them but because that is what happens when your mission requires infrastructure someone else owns. That is a lesson about institutions rather than about AI and it generalises.
It won the Financial Times and Schroders Business Book of the Year for 2024, which was the right call in a strong field. The limitations are mostly about timing. Published in 2024, it stops before a great deal that has happened since, and anybody reading it now needs to supplement it for recent events. The technical explanation is deliberately light, which is a defensible choice for a general audience and does occasionally leave the reader without enough understanding of what a capability actually is to judge the significance of a moment being described.
The thriller framing, which the marketing leans on, occasionally sharpens motives past what the reporting can support. Real people making decisions under uncertainty are messier than characters in a race, and the book is at its best in the passages where it lets them be. The geographic narrowness is the gap I would most want filled. This is a story about two Anglo-American organisations and their corporate parents.
The Chinese labs barely feature, European efforts get little space, and the picture of a global technological transition that is really a duel between two founders is incomplete in ways that have become more obvious since publication. Four point one. Original reporting, a genuinely important structural argument about what happens to mission-driven organisations that need somebody else's compute, and the best narrative account available of how the current landscape formed. Marked down only for its publication date and a frame that is narrower than the subject.