Amazon и Google: гонка на 200 миллиардов (и кто выживет?)

Imagine deciding to spend two hundred billion dollars in a single year. That is not just a lot, it is more than the annual GDP of Hungary or Kuwait. That is exactly the bet Amazon is making for 2026, while Google is trying to keep up by budgeting as much as $185 billion. We have officially entered an era in which the entry ticket to the club of AI leaders costs so much that even very wealthy countries cannot afford it. But before we start counting other people’s money, it is worth understanding why these figures became reality precisely now and what, exactly, these astronomical sums are being spent on.

Just a couple of years ago, the capital expenditures of tech giants looked substantial but predictable. Companies were building cloud storage infrastructure, upgrading fiber-optic cables, and buying standard servers to support search and parcel delivery. Everything changed with the arrival of generative AI. It turned out that the old infrastructure simply could not handle the new tasks. To train GPT-5-class models or develop in-house chips, you need not just data centers but entire cities of compute capacity. Amazon and Google realized that if they do not build these “knowledge factories” today, tomorrow they will have to rent them from competitors at twice the price.

Most of this money ends up with NVIDIA and other chipmakers, but that is only the tip of the iceberg. Vast sums are also being poured into solving the energy problem. Modern AI clusters consume so much electricity that Big Tech companies are having to buy stakes in nuclear power plants and invest in small modular reactors. This is no longer just a software race; it is a struggle for the planet’s physical resources. Google understands that its dominance in search is under threat if it cannot make Gemini fast and cheap enough. Amazon, meanwhile, sees this as a chance to cement AWS’s leadership by making its platform the only place where startups can train serious models.

However, this coin has another side, one that makes Wall Street investors nervously adjust their ties. The problem is that, so far, no one can clearly explain when these investments will start generating comparable returns. We are seeing colossal Capex spending, but revenue from AI services is growing much more slowly. That raises a reasonable question: are the giants building giant monuments to their own unrealized hopes? If it turns out in two years that the market does not need that much compute capacity, we will face the biggest overcapacity crisis in the history of the IT industry.

Even so, stopping now would mean admitting defeat. In this game, there is no second place. If Google decides to save money and falls a year behind in infrastructure development, it could lose forever the search market that fed it for decades. Amazon is in a similar position: its marketplace generates pennies compared with cloud services, and losing AWS leadership would be fatal for the company. So they will keep burning billions, hoping that their architecture becomes the standard for future AI.

Key point: We are witnessing the greatest restructuring of the global economy, where the measure of success is no longer code but the number of gigawatts and teraflops under the control of a single corporation. Will they be able to turn this “hardware” into real money before investors run out of patience?