There is something almost paradoxical about the AI revolution. On one hand, it is the most influential technology of recent years. It has already changed the way we search for information, develop software, conduct research, manage teams, create content, and much more. On the other hand, the companies driving this revolution are spending enormous sums on it, losing millions every day, and burning cash at a rapid pace.
The gap between AI’s massive impact on the world and the economics behind it is one of the most interesting financial stories of our time. In the following lines, we will try to understand why this is happening, and what we can expect next.
A Real Revolution, an Enormous Bill
To understand the troubled economics of the AI revolution, let’s start with the numbers behind some of the most important companies in the field.
Alphabet, Google’s parent company, expects its capital expenditures in 2026 to reach $175–185 billion. According to the company, a significant portion of this budget will be dedicated to expanding its computing capabilities, building and operating server farms, data centers, and the networks required for AI. Amazon expects expenses of around $200 billion in 2026, and it too is expected to dedicate a significant portion of that budget to AI infrastructure. Meta expects capital expenditures of $115–135 billion in 2026, and Microsoft announced as early as January 2025 that it expects to invest about $80 billion in “AI-enabled” data centers for model training and the deployment of AI and cloud applications.
It is important to be precise: not all of these sums are dedicated directly to AI. But all of the giants are stating that a significant percentage of them will go toward the infrastructure AI depends on: the nonstop production of chips, enormous amounts of electricity, cooling systems, the real estate on which data centers will be built, salaries for infrastructure engineers, research, and more.
It is also important to say that AI costs do not stop at the development stage. As more people use AI systems, the cost only grows. Every query or prompt a user submits costs money for the AI developers behind the system. In the coming years, operating costs are expected to reach trillions of dollars.
The AI Race: The Real Cost of Competition
The problematic cost of developing and operating AI infrastructure is also heavily influenced by the fierce competition in the field. In most cases, high costs lead to competitive restraint. In the AI market, the opposite is happening. Every leap made by one player pushes the others to respond, whatever the cost may be. And so, a race has begun in which no one wants to be left behind, even if that means massive investments before the business model has fully stabilized.
The economic gap of the AI revolution becomes especially clear when looking at the two companies that have become its symbols. OpenAI reached annual revenue of $20 billion in 2025, but it is still not profitable. It has raised about $60 billion to date, and according to forecasts, it is not expected to show profits before 2029 – a year in which it hopes to reach approximately $100 billion in revenue and around $14 billion in profit. By the end of 2028, its cumulative losses may reach $44 billion.
Anthropic, the developer of Claude, is moving in a similar direction. It jumped from an annual revenue run rate of $19 billion at the end of 2025 to $30 billion by the end of March 2026, but that growth relies on massive fundraising: $30 billion at a valuation of $350 billion in February 2026, followed later by a valuation of $380 billion. This comes alongside reports that investors were willing to invest in the company at a valuation of $800 billion.
In other words, both companies are growing at an almost unimaginable pace – but that growth is based on massive fundraising and a promise of profitability that still lies far in the future.
A Revolution on Shaky Ground: The Real Challenge of the AI Revolution Is Not Technological, but Financial
The possible solutions to the troubled economics of AI applications will probably not come from one dramatic idea, but from a deep improvement in efficiency. Some are already talking about highly ambitious directions, such as computing infrastructure in space, but for now these are more experimental ideas than proven solutions. The more practical path lies elsewhere: dedicated chips, better energy utilization, more efficient data centers, models that can deliver more value with fewer resources, and of course, more precise pricing architectures from AI providers.
If the first generation of AI was built mainly on brute force – more capital, more electricity, more capacity – the next generation will have to be built on smarter economics. Whoever leads the next stage of this revolution will not necessarily be the one who builds a stronger model, but the one who manages to operate it at a more reasonable cost and within a sustainable business structure. The history of technology is full of brilliant innovations, but those who truly survive are usually not only the ones who invented the future, but the ones who managed to finance it without collapsing under its own weight.