Empire of AI Summary: Who Really Controls AI? (Karen Hao)
Prefer to watch? Open the animated summary on YouTube.
In one lineAI companies behave like empires: they take data, labour and resources from weaker places and gather the rewards at the top.
Key lessons
- AI is not only code. It is a physical, political and human system.
- OpenAI moved from an open, nonprofit mission to a secretive, commercial race.
- Hidden workers, such as data labellers and content moderators, make AI tools look safe and smart.
- Data centres use large amounts of power and water, often in places with little say.
- Belief in 'inevitable' super AI is used to justify secrecy and concentrated power.
- The current path is the result of human choices, so it can change.
- Ask whose data, whose labour and whose land sit behind the tool you use.
On this page (8 sections)
You type a question into a chatbot and an answer appears in seconds. What you do not see is a worker in Kenya who read very upsetting text so the tool would stay polite. You do not see the data centre drawing water and power from a dry region. You do not see the annotator, perhaps in India, who labelled thousands of images for a small fee.
Empire of AI by Karen Hao is about this hidden side. It asks what the rise of AI means for people who want to build a career and earn well. It also asks how they can do that and still feel good about it.
What is Empire of AI about?
Karen Hao is a journalist who has reported on AI for years. For this book she had unusual access inside OpenAI, including interviews with many current and former staff. The full title is Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI.
OpenAI is the main character, but the real subject is bigger. A small group of well-funded tech leaders is building a new kind of empire. Hao’s core idea is that AI companies act like historic empires. They take resources from weaker regions and gather wealth and power at the top. They justify it with a big promise: advanced AI will one day save humanity.
This matters to you because the empire depends on your data, on the job market and on public rules.
How an idealistic lab became a powerful company
OpenAI began in 2015 as a nonprofit. Its stated aim was to build safe AI and share it so that no single company or government would control it. Idealistic researchers and donors joined for that reason.
Hao describes an early bet on what she calls the scaling approach. The belief was that bigger models, more data and more computing power would lead to smarter systems. That sounds technical, but it has a large cost. You need data from across the internet, huge amounts of power, water to cool machines, and many people to clean and label data.
A nonprofit that shares everything could not keep up with that hunger. So OpenAI created a capped-profit company, took billions from Microsoft and raced to release ChatGPT. On paper the mission stayed. In practice the behaviour changed.
Where the hidden costs fall
An empire is not just rich. It gains from unequal relationships. Hao follows the AI supply chain from top to bottom.
At the top are boardrooms in San Francisco and talk of benefits for all. At the bottom are moderators in Kenya paid low wages to review disturbing content. There are also data workers in places such as Venezuela and the Philippines doing repetitive labelling.
India is part of this chain too. It is already a large hub for data labelling and support work. The risk, in Hao’s view, is that rules written only in Silicon Valley could leave India with the low-margin, high-stress end of the work. Profit and decisions would stay abroad.
The costs also include land and water. Hao points to places like Chile, where communities that already face water stress must compete with new data centres. Companies often choose sites where rules are weak and power is cheap.
Belief as a business tool
Inside OpenAI, Hao describes two camps. Optimists expect an age of plenty. Pessimists fear disaster. They disagree on the outcome, but they agree on one thing: very powerful AI is coming, and whoever gets there first wins.
There is no clear definition of artificial general intelligence, so the finish line is always just far enough away. That lets leaders justify almost any step. They say they need more data, more secrecy, more money and less regulation, because rivals might get there first.
Hao’s view is that this is closer to faith than to science. It persuades investors, staff and even lawmakers to accept an extreme concentration of power.
Who controls the story
In its early years OpenAI spoke often about openness. As the race tightened, data sets and training methods became private. People who raised hard questions were sidelined or left.
Hao links this to November 2023, when OpenAI’s board removed Sam Altman and brought him back within days. Staff protests, investor pressure and Microsoft’s support all played a part. Her reading is blunt. The board was meant to put humanity first, yet the people with capital and infrastructure held the real power.
Three tough truths
First, the current path of AI is not natural progress. It is the result of many choices made by a small group with financial and political motives. “We had no choice” is a sales line, not a fact.
Second, “AI for all” does not describe how value is shared today. The gains go to a few firms and investors. The stress, surveillance, unstable work and environmental cost spread across many people with little say.
Third, silence is not neutral. When researchers who ask about bias, safety and fairness are pushed out, the public gets a managed story.
How to use this at work and at home
- Ask where your company’s data comes from, who labels it and under what conditions. Treat low pay for harmful work as a warning sign.
- When you design a product feature, trace it downstream. Ask who could be harmed, whose job changes, and what power or water it needs.
- If you invest in AI firms, look past the hype. Prefer those that are open about data sources, pay fair wages to contractors and report their environmental impact.
- Use AI tools, but ask on whose terms. Support clear rules at work on data use and opt-outs.
- When you hear “light-touch rules so innovation can grow,” ask: innovation for whom, and paid for by whom?
Our take
The book’s strength is its reporting. Hao gives names, places and working conditions, which makes a complex topic feel real. Her view that AI is a system of power, not a neutral tool, is useful and rarely heard on the business side.
The weak points are fair to note. The tone is critical throughout, and the book gives less space to what AI tools do well for ordinary users and small businesses. The empire comparison is a strong frame, but it can make a mixed picture look one-sided. Readers should also remember that any account of a private company rests on sources who may have their own views.
Still, if you use AI daily, this book adds a view you are unlikely to get from product announcements.
The animated summary walks through each idea with stories and examples. It takes about 13 minutes.
Who should read Empire of AI?
Engineers, product managers, investors and curious readers who use AI every day and want to understand the people and costs behind it.
Watch the full animated summary
Every idea from Empire of AI, told as a story in about 13 minutes.
As an Amazon Associate, we earn from qualifying purchases.
