AI efficiency doesn’t always lead to economic growth

by Fedorov Mikhail

At the onset of 2026, investors who gaze intently upon the technology sector are experiencing not merely doubts, but, I would say, a deep and strange feeling of cognitive dissonance. The situation is paradoxical. On the one hand, purely technological progress is absolutely indisputable: models have become faster, they have become smarter and, what is not unimportant for business, they have become significantly cheaper in exploitation. On the other hand, dry financial reports of tech giants and the nervous reaction of the stock market are beginning to demonstrate visible fatigue. That insane euphoria of 2024–2025, when any startup with the prefix “AI” received money by the sackful, is gradually being replaced by anxious, uncomfortable questions: “But where, properly speaking, is the money? Where are those trillions of dollars of net profit, which were so confidently promised to us?”

The answer to this question, by my deep conviction, lies by no means in the program code and even not in the quantity of parameters of new neural networks. It lies in boring, but inexorable macroeconomics. To truly understand the future of the market and, what is more important, avoid painful losses in the oncoming period of turbulence, we must introduce a clear, rigid coordinate system. We must divide the influence of AI on the economy into two fundamentally, principally different categories: Type 1 (Optimization and Substitution) and Type 2 (Creation of fundamentally new value).

The confusion between these two notions — is, perhaps, the main reason why many investors right now risk finding themselves in a trap of inflated expectations.

Chapter 1. Theory of Value: The Illusion of GDP Growth

The key to understanding the current situation lies in how exactly AI influences aggregate demand and aggregate supply. Let us disassemble this at the basic, “molecular” level.

Type 1: Optimization (The Efficiency Trap) The first type — is the usage of AI in order to make already existing things better, faster, or banally cheaper. This is the automation of routine, improvement of logistics, optimization of advertising algorithms or, direct replacement of human labor with program code.

In this scenario, AI acts as a “smart assistant” or as a full replacement for a worker. But here there is a macroeconomic catch, a trap, which can be called the Paradox of the Slipper Factory.

Let us conduct a thought experiment. Imagine a world which stably consumes 1 billion pairs of slippers per year. Demand for slippers is inelastic — people do not need 2 or 3 billion pairs only because they became easier to produce. People have only two feet. If you introduce AI and robots to this factory, you, undoubtedly, can fire half the personnel, cut costs on the payroll fund and sharply increase the marginality of the business. This is excellent for the profit of a specific company (be it conditional Amazon, Walmart, or Nike). Shareholders are happy. But from the point of view of world GDP, the pie has not increased by an inch.

• Before: The cost of slippers in the economy was composed of workers’ wages + owner’s profit.

• Now: The cost is composed of server rent/software licenses + increased owner’s profit.

What happened? A simple redistribution of income occurred (from labor to capital and algorithms), but aggregate final demand in the economy did not grow. The world did not become richer, it simply became more effective in the production of the old. This is a classic zero-sum game on the scale of macroeconomics. And precisely in this phase — Type 1 — we, for the most part, resided for the last two years. We mainly optimized the old.

Type 2: Creation (Creation of Value) The second type — is a completely different story. This is the usage of AI for the creation of goods and services which physically did not exist before the appearance of the technology. These are products which do not replace old consumption, but create new, expanding the boundaries of the market.

Recall the appearance of the iPhone and App Store in 2007–2008. After all, this was not simply an “improved telephone” or “telephone with a touch screen.” No. This created a gigantic industry of mobile games, taxi services (Uber), instant food delivery, and social networks right in your pocket. People started spending money on things on which they previously did not spend, simply because they did not exist in nature. This is an expansion of the nomenclature of goods, this is real, organic GDP growth both by demand and by supply.

In 2026, the AI Market, in my view, will abut into the concrete ceiling of Type 1. We gathered almost all “low-hanging fruits” of optimization — cut copywriters, optimized code, improved search. And now, for the continuation of growth, we vitally need a painful, but important transition to Type 2.

Chapter 2. Mirage of Autonomous Agents

One of the hottest, I would even say, “scorching” themes of late 2025 — early 2026 became the so-called “Autonomous Agents.” Technologically this is, undoubtedly, a step forward: instead of chatting with a chatbot, you give a task to an agent, and he fulfills it (books tickets himself, consolidates tables himself, writes and tests code himself).

Many call this a revolution. I, however, urge investors to be more cautious and take off rose-colored glasses. From a dry economic point of view, Agents — are still Type 1, simply in very beautiful, shiny marketing packaging.

Let us be honest with ourselves: if an AI agent replaces an assistant accountant or a tourism manager, what really happens with GDP?

• Substitution Effect: Work, which a live human did, is now performed by soft.

• Cost Structure: The company pays a subscription to Microsoft or Salesforce instead of a salary to an employee.

• Result: The quantity of accounting reports or booked tours in the world did not increase. There are just as many.

Agents — are a most potent deflationary factor. They lower the cost of doing business. Yes, this will increase the profitability of corporations in the short-term perspective (what is called margin expansion), but this is not a driver of organic growth of the economy. Moreover, mass introduction of agents can lead to pressure on the labor market, which, in a paradoxical manner, can lower consumer demand. The fired accountant banally has no money for new goods. Therefore one should not confuse a technological breakthrough (Agent works himself, hurrah!) with an economic breakthrough (Market grew!). While agents engage in substitution, they remain in the trap of Type 1.

Chapter 3. Factor 2026: China, DeepSeek and “Race to the Bottom”

In my last year’s analysis on Seeking Alpha, I, I confess, did not account for the Chinese factor in full measure due to a lack of verifiable data. In 2026, to ignore it — is already criminal professional negligence for any analyst.

The appearance and aggressive scaling of models of the DeepSeek type (and their analogues) changed the rules of the game inside Type 1. They proved a terrible thing for American Big Tech: efficiency (intelligence of GPT-4/5 level) can cost several times, or even orders of magnitude cheaper than is customary to consider in Silicon Valley.

What does this mean for the market? This signifies the commoditization of intelligence. A smart model ceases to be a unique, sacral advantage of Google or OpenAI. It becomes just as accessible, utilitarian a resource as electricity in a socket or internet traffic.

• If intelligence is cheap, then the margin of companies selling “simply access to the model” will inevitably fall.

• A price war begins. Cruel and merciless. Competition in the Type 1 segment (optimization) turns into a race where the margin goes down.

This creates colossal, as yet underestimated pressure on American tech giants. Their business models, built on the sale of expensive subscriptions to the “smartest AI,” turn out to be under threat of dumping from the side of open source and Chinese super-optimized architectures. This is yet another signal, a loud bell that the epoch of easy money on Type 1 is ending.

Chapter 4. Risk of Nvidia and “Investment Phase”

We approach the most interesting, most acute moment for the stock market. If Type 1 has limits (and we see them), and Type 2 has not yet really gathered mass, on what, damn it all, did the market grow in 2024–2025? The answer is simple: It grew on CAPEX (Capital Expenditures).

We are located in a unique historical point, which can be called the “Investment Phase.” Hyperscalers (Microsoft, Google, Amazon, Meta) spend hundreds of billions of dollars — insane, unimaginable money — on the construction of infrastructure, data centers, and procurement of chips.

For Nvidia, this is, undoubtedly, a golden time. Nvidia sells “shovels” during a gold rush, and does it virtuously. Technically, GPU production — is Type 2 (this is a new good, which previously was not needed in such quantity). Demand for GPUs — is indeed a new market. But here lies the “Derivative Trap”.

Demand for Nvidia chips is a derivative of the hopes of their clients (Microsoft, Google) to earn on AI. The chain looks thus:

• Nvidia sells a good of Type 2 (chips).

• But her clients utilize these chips for now mainly for tasks of Type 1 (optimization of search, advertising, coding).

If in 2026 the CFOs of Google and Microsoft realize that the economy of Type 1 does not scale infinitely (one cannot optimize search to infinity, one cannot fire all employees), they will start to cut costs. And the first to go under the knife will be the procurement of new hardware.

Consequently, the current cosmic valuation of Nvidia — is a bet of the market on the fact that the transition to Type 2 will happen before demand for optimization ends. This is a race against time. If new sales markets do not appear in the nearest year or two, we risk seeing a classic cycle of infrastructure overproduction (overbuild), exactly as it was with the laying of thousands of kilometers of fiber optics during the dot-com crash. The cable lay, but there was no traffic for it.

Chapter 5. Where to Seek the Holy Grail (Type 2)?

To criticize, of course, is easy. But whither to look for a reasonable investor? Where is located that selfsame Type 2, which will save the market, restart the growth cycle, and justify trillion-dollar investments? We need to seek products which create new pastime or new physical possibilities. That which expands the boundaries of human experience, and not simply saves us 5 minutes of time.

Here is my list of directions, which can become a real Type 2:

1. Market of Cloud Computing as a Historical Prototype It is worth noting for context, that Cloud Computing itself was a successful example of Type 2 in the past decade. It created infrastructure on which grew Uber, Airbnb, Netflix. Without clouds, these businesses would be physically impossible. Now AI must become just such a platform for something fundamentally new.

2. Chatbots and LLM as a New Service Sector (The Second Type) And here we must make a very important addition. We often scold Chatbots, but let us look closer. The very communication with a smart machine — is a service, which did not exist before. Previously, you could not pay $20 and receive an interlocutor who knows everything, writes code, and helps to think. This is not optimization of old search (though Google tries to present it so). This is a fundamentally new market of services.

• For Google and OpenAI, ChatGPT or Gemini — is not just a “better search engine.” It is a new product category.

• Millions of people now pay for a subscription. They did not pay for search before. They pay for intelligence on demand. This is the first, earliest sprout of Type 2, which has already sprouted and brings real money. It is a new market of sales, created from scratch.

3. Hyper-personalized Entertainment Content Streaming (Netflix) in its time killed cable TV, but this is still a mass product (“one film for all”). AI of Type 2 — is the generation of a film, game, or virtual world personally for you, in real-time. This is not simply “content”, this is a new market of struggle for dopamine and the free time of a human. This market does not exist now, its volume equals zero, but its potential — is huge.

4. Robotics (Physical AI) While AI lives in a chatbot on a screen, it is limited by the world of bits. Exit into the world of atoms — is a real Type 2. A robot (for example, Optimus from Tesla or analogues), which can work 24/7 in dangerous conditions, extract resources on asteroids, fix pipes on the ocean floor or care for elderly people — this is the creation of a new labor force, which previously did not exist. This is not the substitution of a conditional loader, this is the possibility to do projects which previously were economically inexpedient due to the high cost of labor. But here we abut into heavy technological limitations of “hardware,” which are harder to overcome than simply writing code.

5. New Science and Biotech Usage of AI for generation of new materials or medicines. If AI invents a cheap cure for aging or helps to stabilize thermonuclear fusion — this is Type 2 in pure, distilled form. This is a good for which humanity will give any money, and which previously simply was not in the price list of the world economy.

Chapter 6. Analysis of Key Players: View 2026

Proceeding from this logic, let us try to soberly re-evaluate the giants.

• Google (Alphabet): The company, in my opinion, got stuck in the classic “innovator’s dilemma.” Their main income (Search) — is Type 1. Introduction of AI thither merely cannibalizes their own margin (generation of a smart answer costs more expensive than showing ten blue links). They vitally need to develop Chatbots as a separate paid product (Type 2) and find other new markets, otherwise they risk becoming simply a very expensive “infrastructure pipe.”

• Microsoft: Made a huge bet on Copilot (Type 1). Sales go to corporate clients, but reviews, let us say so, are mixed. If they do not turn Copilot into a platform for creating new applications (like Windows in the 90s), growth will slow down.

• Meta: Mark Zuckerberg plays the most cunning, non-obvious game. Making his models (Llama) open, he intentionally destroys the margin of competitors in Type 1. His goal — not to sell AI as a service, but to use it for the creation of the Metaverse (or new social experience). If the Metaverse finally “takes off” with the help of generative AI — this will be pure Type 2.

Conclusion: Investor Strategy

We are entering a phase of sobering up. The hype of the “first wave,” when any company with a mention of AI in the presentation grew by 50% in a month, has passed. 2026 will become a year of reality check. A year of exam.

My recommendations:

• Caution with “Optimizers.” Companies whose sole strategy — is “fire people and install AI,” have limited growth potential (ceiling). This is a game on margin increase, but not revenue. This is a short game.

• Watch for CAPEX. If Amazon and Microsoft continue to blindly increase costs on hardware, not showing commensurate revenue growth from new products, the market will punish them with a rigid, possibly even cruel correction.

• Seek Market Creators. Alpha (super-profit) is now located not in the shares of chip manufacturers (there all expectations are already in the price, and even more), but in those who will invent the “Uber of the AI epoch.” In those who will find application for this excess intelligence for the creation of new added value.

History, as is known, loves to repeat itself. In 1999 we laid fiber optics around the whole world (investment phase), the dot-com bubble burst, investors lost money, but cables remained lying on the ocean floor. And only 5–7 years later on this ready infrastructure grew the real Internet 2.0. Now we are building “neural highways.” Be ready for the fact that trains (Type 2) might not go along them immediately. But when they go — this will change the world stronger than the iPhone.