Andreessen Horowitz has raised 1.1 billion dollars for a new fund called "Machine Age", with the declared intention of "stepping on the gas and accelerating the physical build-out of artificial intelligence". The firm that spent years making money on software is now putting it into iron.

The target is the infrastructure underneath the models: chips, memory, interconnects between nodes, data centres, robots. The fund's announcement carries a wish list written as technical requirements - "faster and more efficient systems", "cheaper memory with higher bandwidth", "more scalable connections between nodes and systems", "energy-efficient devices at the edge of the network". And at the end, almost in passing, the part that actually costs money: cooling, materials, electricity and the construction work that has to hold all of it up.

That last item is the point nobody says out loud. Artificial intelligence is not a cloud - it is a building with power, cooling water and a plot of land. When one of the world's largest venture funds opens a dedicated pot for "physical build-out", that is an admission that growth is no longer limited by ideas, but by transformers and real estate.

The language in the announcement is worth reading closely. AI is described as "the most powerful tool yet developed for solving problems and delivering abundance", and its progress as "a societal and national imperative". That is not investor analysis - that is a political sentence, written by a fund managing other people's billions. When a private investment is translated into a national duty, what usually follows is a request for permits, for a grid connection, and for something out of the state budget.

The European side of the ledger is quieter. Data centres demand power and water, and those are things every country calculates locally, not in California. Who gets the connection when the grid is old - that is a question this fund does not solve, and every municipality in its path will have to answer it.

1.1 billion is a big number for a fund, but a small one for construction. That sum does not build an empire of data centres - it buys stakes in the companies that make its parts. It is a more sensible bet than the headline sounds: instead of guessing which model wins, you invest in the fact that all of them will need cooling.