Karen Hao: "Big Tech Wants to Privatize Welfare Too"

Karen Hao is the journalist who, with Empire of AI, has rewritten the way we read about OpenAI and the frontier model industry. At AI Week 2026, she took the Main Stage for a keynote that centered around a word that AI companies would prefer not to hear, namely "cost." Immediately afterward, I intercepted her offstage for a few questions. I report here the public speech and the private conversation, as only by reading them together can we fully grasp her thesis.

The Keynote: An Empire, Not a Company

Hao's starting point is that OpenAI should be described as an empire rather than a tech company. It’s how she interprets the combination of economic scale, political influence, and consumption of physical resources that characterizes this sector.

To explain this, she started from the most uncomfortable side of model training. She talked about the work of those who filter violent content used for training, people who spend hours viewing hate speech material, scenes of violence, child abuse, so that the model learns to recognize and not reproduce them. It is a job done by humans, and Hao used it to pose the question that runs through her entire intervention: do we really need empires of this size?

From here, the keynote shifted to physical infrastructure, and this is the part that struck me the most, as it was full of numbers and specific locations.

Water, Land, Missed Innovation

Hao compared the dimensions of next-generation data centers with the layout of New York City to illustrate what we are building. Meta's upcoming data center, called Hyperion, under construction in Louisiana and expected for 2028, is estimated to cover an area of eleven square kilometers. Not far away is OpenAI's Stargate in Texas and xAI's Colossus 1 in Memphis. Structures as large as neighborhoods.

The point she insisted on, and that she presents in her book, is that this scale is not necessary. It is not an inevitable technical consequence; it is an industrial choice, and according to Hao, it would be possible to create useful AI without reaching these dimensions.

Karen Hao on the Main Stage at AI Week 2026.

Then she moved on to water. She presented data on the increasing number of data centers built in areas already under water stress, in the United States and around the world, with a curve that steepens after the release of ChatGPT.

She recalled a Guardian investigation on the Big Tech data centers placed in the driest areas of the planet, and the stance of Lorena Jaume-Palasí, founder of the Ethical Tech Society, who stated that the combination of the climate crisis and expansion of data centers is bringing Spain to the brink of ecological collapse. I was struck by the detail of an Amazon data center requesting a 48% increase in its water consumption.

Hao argued that all this is costing us innovation. She recalled two connected facts: two AI giants based in San Francisco that alone raised almost half of all venture capital in one quarter, and the 40% collapse in climate tech funding in 2024, as investors flocked to artificial intelligence. The concentration of capital on very few AI players is not only consuming water and energy but is also draining resources that would have been needed for other technologies, including those that address the climate crisis.

The Power of Citizens

At the end of her speech, Hao turned the perspective around, and this is the transition to my questions. When we think about our relationship with AI, we see ourselves as the last link in the chain, as users of a finished product. She argues that we are part of the supply chain at many more points. We are workers whose work can end up in the training of models. We are data providers. We are citizens of a community where a data center may land.

From this interpretation comes her thesis about the margin of action for people. Around the world, Hao says, communities and organizations are realizing they can have an impact. She cites protests against data centers and the fact that in 2025, over one hundred billion dollars in projects were blocked or slowed down due to these oppositions. The data holds: according to Data Center Watch tracking, in 2025 alone, projects worth about 156 billion dollars were halted or delayed, with the second quarter alone valued at nearly one hundred billion. She adds lawsuits from authors and publishers over intellectual property theft, lawsuits from parents for the psychological harm suffered by their children, and consumer boycotts.

The operational point is that everyone can choose whether to accept or reject what happens in their neighborhood, at work, in their children's school, in their doctor's office. AI is not a fate imposed from above.

First Question: Sam Altman's UBI

Backstage, I asked her, from a European perspective, how we should interpret the universal income proposals coming from Silicon Valley CEOs. In her book, Hao traces the thread linking Altman's public advocacy for UBI to OpenAI's industrial strategy as a provider of automation. I wanted to know whether those proposals should be seen as good faith, as preemptive lobbying to legitimize job destruction, or if there is a third key that we Europeans are not grasping.

For those unfamiliar with the acronym: UBI stands for Universal Basic Income, the idea of guaranteeing every citizen a periodic sum regardless of the work they do.

Hao says that the second interpretation is correct, and there is also a third, more disturbing one. Traditionally, what companies describe as UBI is the responsibility of the State, because it involves a social safety net managed by the government. What these companies are saying, according to Hao, is something else: to dismantle the public safety net and replace it with a system built by them, devoid of any form of accountability to citizens. These companies are effectively proposing to decide if a person deserves a safety net or not. They are trying to privatize what should be a public good, just as they are doing with many other things meant to be public.

I pause for a moment here because Hao touches on a thread I’ve been following for months (I also discussed it in chapter 11 of my book Umani per Ora). One thing is to discuss whether UBI is sustainable from a macroeconomic standpoint, and that is a legitimate discussion. Another is realizing that the true stakes, in Hao’s reading, concern who holds the levers of control. If the safety net passes from a public institution, which at least in theory answers to voters, to a private company that answers to no one, then universal income ceases to be a guarantee and becomes a revocable concession. The same mechanism that Hao denounces for water and land, namely a common good that is being privatized, she finds in the welfare proposal.

Second Question: Does European Regulation Work?

I then asked her about governance. In the book, Hao describes how OpenAI has become a political actor more than a technological one, an entity that influences policies more than it develops technology. I wanted to know if she sees European regulation as a tool to prevent the same governance failures documented in the United States, or as a ballast that hinders the emergence of a European "AI empire."

Here, the answer turned a common notion upside down. Hao sees it as a positive sign that Europe does not have an AI giant resembling the American behemoths. For her, this means that regulation has worked, that it has managed to break the concentration of political and economic power in the hands of a few companies. The absence of a European OpenAI, in her view, is proof of effectiveness.

However, there remains a knot that Hao raised in both answers, and it is the point worth closing on.

What Europe Lacks

What Hao believes Europe lacks is a strong vision of an alternative model for AI development. Her argument is that we need to know not only what we are fighting against—the tech empires—but also what we are fighting for. What version of AI can we all truly rally behind? A sustainable version that values people’s freedom of action and supports workers' rights and human rights.

And here comes the operational part. When Europe clarifies what model it wants and directs real funding towards it, Hao predicts that the entire ecosystem of AI technologies will start to shift away from the more extractive and exploitative versions.

It is worth pausing here, because it is a constructive critique.

The day before this conversation, I heard Lucilla Sioli, director of the EU AI Office, speaking about how Europe is building its AI factories and giga factories. Regulation is in place; the AI Act is being implemented; the infrastructure is being constructed. What Hao points out as missing is a shared vision that clearly states what type of artificial intelligence Europe wants to bring into the world, and why.

Humans, For Now

There is a reason I chose to title my book "Umani per Ora" (Humans, For Now), and it has to do with that "for now." It is not a melancholic jest about the impending end of something, but rather an indication of time. It means that right now, we are still the ones who can decide what kind of artificial intelligence we want, where the data centers will be located, whose water they consume, and who manages the safety net when work changes. Hao has expressed it clearly: citizens have more power than they think, and in many places around the world, they are already using it.

The point is that this window will not remain open forever. The day a private company decides on its own whether a person deserves income or not, the possibility of choice will have already passed from hand to hand. As long as the question of "what AI do we want" remains an actual question, and not a pre-written answer in an office in California, the "for now" in the title continues to hold value. Keeping that window open and telling readers it exists is a big part of why I write.