A $650 Billion Arms Race: Big Tech Bets Everything on Chips and Concrete

When the numbers climb past half a trillion dollars, ordinary arithmetic stops working and what begins is corporate-scale geopolitics. Amazon, Microsoft, Google and Meta have decided that modesty is not for them, and have budgeted a combined $650 billion in capital expenditure for this year. To put that in perspective: the sum exceeds the annual GDP of most European countries. This money will not go to new interfaces or marketing campaigns. It will go into concrete, copper and, of course, silicon. We are watching the largest rebuild of the technological base in human history, in real time.

Just a year ago, the industry was debating whose language model tells better jokes or writes code faster. Today the discussion has shifted decisively to the physical world. It turns out the "magic" of artificial intelligence requires very tangible and very expensive things: giant hangars, millions of miles of cable, and as much electricity as entire cities consume. If cloud technology once seemed ephemeral and intangible, it is now the heaviest, most resource-hungry industry in the world. The companies are literally burying money in the ground, laying foundations for future neural networks that, they hope, will someday justify the investment.

It is striking how sharply these giants' rhetoric has changed. Where investors were once promised optimization, buybacks and cost cuts, they are now told bluntly: we will spend a lot, a great deal, and you will have to live with it. Mark Zuckerberg, who not long ago preached a "year of efficiency," now talks eagerly about buying hundreds of thousands of GPUs. Microsoft and Google are keeping pace, aware that in this race for compute, second place means oblivion. It is the classic prisoner's dilemma, scaled up to Big Tech: if you don't build a data center today, tomorrow your competitor trains a model you have no chance against.

Amazon, through AWS, has always been the king of the cloud, but now it has to run twice as fast. It is not just building server farms; it is designing its own chips to cut, even slightly, its dependence on outside suppliers and Nvidia. This is vertical integration on steroids. When you spend tens of billions per quarter, saving even five percent on hardware costs turns into enormous sums that can fund yet another facility. The entire hardware market is now working to satisfy the appetites of these four, creating component shortages for everyone else.

Behind this financial optimism, however, lies a very concrete problem: hunger for power. Constructing a building and buying servers is only half the job. They must be powered, reliably and in huge volumes. We already see tech companies buying stakes in nuclear power plants and investing in small modular reactors. The AI world has suddenly run into the hard limits of physics. Training the next generation of models, it turns out, is not only a question of algorithms but also of whether the local grid in Virginia or Iowa can handle the load. The environmental agenda has temporarily taken a back seat to technological dominance.

Wall Street watches this parade of generosity with a mix of delight and quiet horror. On one hand, reports show growing revenue from cloud services. On the other, free cash flow looks less attractive when capital spending eats almost all of it. Analysts are asking when the moment will come that AI services bring in as much as the companies spend to sustain them. So far we see only rising resource consumption, while real monetization for mass consumers is still at the stage of experiments and paid subscriptions.

What does this mean for the market as a whole? Most likely, even harsher monopolization. With an entry threshold in the hundreds of billions of dollars, new players and startups can only hope for the giants' mercy or try to build something remarkably efficient on modest compute. The era of "garage" innovation in large models is finally giving way to an era of industrial gigantism. We are entering a period in which AI success is measured not only by code quality but by terawatt-hours and the floor area of server racks. The winner will be whoever has enough money to finish the last data center.

The takeaway: Big Tech is turning the software industry into heavy machine building. Can it monetize these costs before investors run out of patience?

*Meta has been designated an extremist organization and is banned in Russia.