The neural network revolution promised humanity more free time, rising intelligence, equal opportunities, and the disappearance of intrusive advertising. Four years have passed since ChatGPT launched — and on almost every point, reality turned out to be the opposite. Each unfulfilled promise, however, points precisely to where the money in the AI economy actually goes and who receives it.
Programmers were the first to be hit
The logic seemed flawless: the more AI there is around, the more people are needed to implement and maintain it. In reality, those who write code were the first to suffer. A study of payment data from ADP covering millions of workers showed that employment among developers aged 22–25 fell by almost 20% from its 2022 peak. The total number of developers at the same companies continued to grow — but at the expense of employees over 30. The decline is strongest where AI replaces people rather than helping them. The winners are companies that cut staff, not those that expand it.
There is more work, not less
They promised that AI would take over the routine and people would rest more. Surveys record the opposite: 77% of employees said AI added work for them. Among 1,250 office workers in the US, 31% saw their workload increase after AI implementation, and only 16% saw it decrease. Almost half of those whose work increased say it at least doubled. The saved time turned into a new output norm: for companies, it means higher productivity and profit; for employees, the same salary with a heavier workload.
Advertising didn't disappear — it moved into the chat
People expected the AI assistant to replace the search engine and bury banners. Advertising didn't disappear — it moved inside the dialogue. OpenAI launched advertising in ChatGPT on February 9, 2026, initially for some free users and Go-tier subscribers in the US. Six weeks later, the pilot reached an annual revenue of about $100 million; in May, an advertiser dashboard with pay-per-click appeared, and then advertising was opened in the UK. For 2026, the company forecasts $2.5 billion in advertising revenue. Ads are selected based on the topic of conversation and past chats — more precise than a banner, but not its end.
Concentration of capital instead of democratization
They promised powerful intelligence for everyone — the result was the largest concentration of capital in the history of technology. In 2025, Microsoft, Alphabet, Amazon, and Meta spent about $410 billion on capital expenditures, and for 2026 they plan around $725 billion — growth of roughly 77% year over year. About three-quarters of this money goes to AI infrastructure: GPUs, servers, and data centers. Only a few companies in the world can train a frontier model; the rest rent the result from them. Profit from AI is collected within a very narrow circle: either you own a share in it, or you pay it.
The skill of thinking has to be retrained
AI was compared to a calculator for thinking. The problem is that with calculators, people unlearned how to do mental arithmetic. About ten minutes of chatting with a chatbot reduces the ability to solve problems independently — the effect is stronger among those who asked for ready-made answers rather than hints. An EEG experiment showed weaker brain activity in those who wrote essays with ChatGPT, and worse recall of their own text. For an investor, there is a practical risk here: making money decisions by fully trusting the model is a bad strategy.
Access exists, control does not
Formally, the chat opens in a browser for anyone. But the user gets not the technology, but a rental of it: they do not control the model itself and its updates, message limits and subscription price, the list of prohibited topics, and the list of countries where the service operates. An app built on someone else's model lives until the next change in terms — its business can be shut down not by competitors, but by the supplier.
A one-person company valued at $1 billion
In 2024, Sam Altman stated that a one-person company valued at $1 billion would soon appear. More than two years have passed — no such company exists. The reason lies in economics, not technology: the better the models, the cheaper it is to make a generic product and the less it is worth. If an app can be built in a week, thousands of people will build it, and a corporation with the same AI will make a thousand such apps. The money stays with those who own computing power, not those who rent it.
Cryptalist analysis: The combination of these data points indicates a stable pattern: value in the AI economy is shifting from users and app developers to the owners of computing infrastructure and models. For the market, this means further concentration of capital within a narrow circle of hyperscalers and growth of their share in the sector's total profit. Investors should evaluate AI assets primarily through the lens of control over capacity and data, not through promises of increased labor productivity.