The problem is not the data centers. The question is what we will have worth putting inside them.
The loudest business story of the summer belongs to Alex Karp, the head of Palantir, whose shares jumped by nearly 30 percent in a single day after he told investors that the future of his industry is what he calls sovereignty. His customers, he wrote, “have declined to become vassal states” of the big AI labs, labs that, as he put it to investors, believe they “deserve to colonize your enterprise.” Strip away the theatre and a man who sells software for a living is saying something simple: the battle has moved from who has the smartest machine to who controls the data the machine thinks with.
You did not need to look to Silicon Valley for proof. That was at the start of August, and before the month was out the same argument had filled public debates far beyond California, with data centers celebrated as “the future of humanity” by some and met with hard questions about their benefit and their harm by others.
Underneath the shouting and the concrete sits a question nobody in this debate has put plainly. Do we build the capacity to run our own intelligence, on our own data, or do we keep renting it from two or three companies abroad? Rent, or own?
The first bargain, and the room that stayed guarded
We have signed a deal like this once already. To be connected, we handed over the photographs, the messages and the attention of our private lives. They moved into someone else’s building, and we accepted, because the service was extraordinary and the price was invisible. One room stayed guarded. The banks, the most supervised institutions we have, kept the financial half of our lives, and that discipline is real: no industry in the economy accounts for its data more rigorously. Although, said honestly, even that guard was partial, because almost every card payment already travels through Visa and Mastercard, so the direction was always outward, we just never followed it.
Now the guardians themselves are being told to grow and transform, and the same AI pressure knocking on every other door is knocking on theirs.
The second bargain is bigger
This time what is moving out is not the photographs. It is the knowledge and the wisdom of our companies: the questions a business asks and the reasoning behind the decisions it makes. And this time there is more on the table than who sees the data: access, because the terms can change in an email, and price, because the meter belongs to the landlord and the repricing has already begun. One large company abroad learned that the terms of its AI subscription had changed, worked out that staying under the new rules would cost it around a hundred thousand dollars a day, and rebuilt its system on models it can run itself. Another spent its entire annual AI budget in four months. The first time, we paid with our privacy. This time, we would be paying with the way we think.
Today almost every company rents, and until recently that was the only rational choice: the model is American, it runs in a foreign building under foreign law, and the price list is written in San Francisco. But the alternative is arriving. AI good enough for serious work can now run inside a company’s own walls, and even Microsoft now sells AI of the newest class in a form that never leaves the building. For twenty years our data traveled to the computing, always outward. Now the computing can travel to the data. That is exactly why countries suddenly find themselves arguing about where to pour concrete.
What owning actually means
Owning means control over three things, your data, the machine that reads it, and your exit, and no flag on a logo supplies any of them. Look at the difference between the cheap subscription your employees signed up for and the contract your company signs. On the cheap one, what your people type can, by default, become training material for the landlord’s next machine, kept for years, under terms you never negotiated. Corporate contracts close most of this, and that is the point: the protection is contractual, and it lives in documents most companies have never read. So ask three questions. Whose machine learns from our data? Where does it physically live? And what do we keep if we walk away?
A building is not a strategy
And when the buildings come, because they clearly will, a powered hall full of servers is a concert hall without an orchestra. The music comes from the players, not the room. The harder question is whether your data means the same thing twice and whether the decisions and the reasons behind them are written down, because that decides if your company could use a machine next door the day it opens or would simply move its mess into a nearer building. It is the same foundation these columns keep returning to. The building we can pour. The way a company thinks is the one thing nobody can import for it.
That is the half no lab can sell you and no council can approve for you.
Whatever governments decide about the buildings, rent or own is decided company by company, and mostly by accident. Renting today is sensible: it is cheap, it is ready, and owning is not. But renting is getting more expensive at the same time as owning is getting easier, and readiness is built now, on the data you already have. You do not need to own anything tomorrow, only to be ready to own on the day renting stops being cheap, or the day the terms change and you learn what a vassal actually pays.
Karp is shouting it. The enthusiasts are promising it. And municipal councils are asking the sober version of it before deciding on data-center initiatives on their territory: what is the benefit, and what is the harm? Before we argue about the building, there is a prior question: what we have worth keeping inside it. Do you own the meaning of your business, or are you renting that too?
