Brain vs AI: The Economics of Energy and Intelligence
We’re witnessing a fundamental shift in how value is created, but the economics underlying this transformation reveal a profound paradox. The human brain operates on 12-20 watts while performing billions of operations per second.
A single AI data center consumes over 100 megawatts - enough energy to power 100,000 homes. One LLM response requires 6,000 joules, while the brain spends just 20 on an entire stream of thought. Biological efficiency is higher by a factor of 9×10⁸ - 2.7×10¹³.
NVIDIA is the core of AI infrastructure: in 2025-2026, it controls 90-94% of the GPU market. Blackwell and GB300 power models for Microsoft, OpenAI, Amazon, and Google. Data center revenue reached $30.8B in Q3 2025. By 2026, AI data center operating costs will exceed $600B with energy consumption of 29 GW. The price of an H100 chip has dropped from $50k to $15k, but energy costs remain colossal.
The economics are stark:
Training a large model costs tens or hundreds of millions of dollars
An average employee costs $60-80k per year
AI is efficient at routine tasks; humans are more flexible with complex ones
Chip deployment is growing 2.25x per year Energy remains the key constraint
The Paradox of Progress
In Bostrom’s book Superintelligence, he describes the moment when a self-improving system will accelerate progress and reduce the cost of mental labor to nearly zero. The first state or company to create such intelligence will gain a strategic advantage, with copies costing only hardware and electricity.
We’re still far from this scenario. Even the most advanced NVIDIA chips are approximately 10¹² times less energy-efficient than the human brain. AGI and superintelligence remain a dream of their creators, although the speed and autonomy of agents, especially in programming and automation, are already impressive.
Yet something profound is already happening. AI doesn’t yet think, but it already operates at such speed that it forces us to reconsider our own potential and our entire economic structures.
The Death of the Integrator Pyramid
For decades, integrators lived by a simple scheme: sell fixed-price projects based on labor hours, and the more people on the project, the more money earned. Projects often stretched for years along with team budgets.
The external KPI was project success. The main internal KPI was utilization: each employee needed to have more than 75% of their time billable.
The perfect pyramidal economy: a partner in a suit at the top, hundreds of juniors at the bottom, and smart people with Excel spreadsheets tracking hours, plans, and utilization percentages in the middle.
AI is breaking this system.
Take a typical SAP project. Previously, to implement a module, a company would hire 30-50-100 people, half of them juniors, working for at least a year and a half. Each step - requirements gathering, testing, documentation, employee training - dragged on for months. The integrator’s revenue grew with every hour, while the client endured this slow process.
Let me paint an ideal picture that already looks realistic with AI. For the same project, you can take 5-7 people: seniors and process specialists, and connect AI agents. Agents run 80% of tests overnight, generate documentation, and check configurations. The final result is ready in 3-4 months, and the project cost drops dramatically.
For the client, this is enormous savings. For the integrator, it’s a revenue collapse if they don’t reconsider their model and learn to do more projects with fewer people.
Obviously, clients are also beginning to understand that it’s profitable to pay for results, not for the number of hours and specialists.
Who wins? Clients and small boutiques with AI, where teams of 5-7 people with agents can work faster and cheaper than traditional giants. Medium-sized businesses gain access to projects that were previously only for large corporations - and in just a few months.
Platforms like Salesforce and SAP are starting to release their own agents and earn where integrators used to earn.
Who loses? Body-shop integrators: hundreds of juniors are no longer needed. Large corporations are too unwieldy to restructure their pyramids quickly. All consulting companies need to make a sharp turn.
Simple conclusion: the $1.5 trillion IT services market will soon stop selling hours and start focusing more on speed of results. Those who can guarantee it will survive, not those who cling to the old pyramid.
Unlocking Creative Potential Through AI-Augmented Development
Recently we had another heated debate about whether a person who didn’t study programming at university can write code.
The classic arguments: garbage code, doesn’t scale, won’t handle the load, can’t go to production.
My recommendation to skeptics is to try vibe-coding for educational purposes: Loveable, Replit, Cursor (here it’s advisable to listen to a couple of short courses so you don’t get stuck in basic settings).
Not perfect, but... AI: immediately writes documentation tests and fixes itself breaks logic into modules explaining what algorithm it’s using suggests refactoring.
Beginning juniors often write worse code simply due to lack of discipline.
Obviously, writing clean code and creating industrial architecture for high-load applications is a separate profession, not a side effect of knowing syntax.
From our observations on adoption: the boldest and most successful experiments (in terms of the wow effect from a prototype solution for the client) come from our business analysts. They know how to more accurately describe the final result for Cursor/Replit and don’t get stuck on errors and code beauty.
Obviously, for industrial high-load solutions, many modules will have to be rewritten from scratch. But to justify business value - this is huge progress.
It’s hardest for strong programmers because their professional skill - finding non-optimal parts of code or errors - diverts attention from the essence. And the essence is in a different behavioral pattern: creating more detailed requirements, templates for testing and documentation.
I also discovered a paradox: the AI topic has begun to absorb a disproportionate amount of top management time relative to commercial effect (I’m no exception). Many people simply started vibe-coding in the evenings and caught that very breath of fresh air - the dopamine from inspiring successes. And simply doing a project the old way, without the new AI methodology, is no longer as cool.
Moreover, due to the low cost barrier of creating software, many talented people have started building their own products and are showing unprecedented commercial success in record time and time to market. And then they infect everyone around them with this virus.
The Exponential Unlocking of Human Creativity
Technologies in the field of AI have significantly outpaced people’s ability to perceive and implement them. From my observations, AI is moving extremely slowly and without particular enthusiasm, especially in large companies. It’s like compound interest: everyone understands its power, but few make investments regularly or apply it at all.
Marketers and programmers, while seeing the benefits, often subconsciously devalue AI, fearing the loss of their craft skills’ value. Also, experiments with new technologies require patience for experiments that often seem absurd at first glance, but can ultimately change companies’ principles of interaction with the world and business models.
For example: writing programs by talking to a computer, finding the right words to persuade it to make code more efficient; creating new genres in content, experimenting with digital twins and mixing the context of real news and events with fantasies; creating video content on a previously unimaginable scale based on your own avatar and recorded thoughts - all of this is available right now, but almost no one around me even tries to use it.
In the corporate world, there’s a lot of hype about the potential replacement of routine operations and unnecessary actions through autonomous agents or bots, which has essentially been happening for several decades even without LLMs. My opinion is that the main potential of AI is in accelerating creativity in creating new business models and ways of conveying ideas.
Imagine having the ability to “hire” hundreds of artists, designers, or programmers without entering the labor market and seeing the first results and money in your account from real clients within a day. AI gives the ability to quickly test ideas, generate products, and receive feedback from target audiences.
The real potential of AI is in freeing the creative energy of smart people. AI is already turning obsessions and random insights into reality - that is, essentially, it’s primarily an exponential unlocking of human creativity that we didn’t even suspect existed.
The Energy Economics: A Constraint and an Opportunity
Yet this democratization of creativity comes with a fundamental cost. While AI accelerates development and unlocks human potential, it does so at an enormous energetic price. The $600B in projected data center operating costs by 2026 represents not just a financial burden but a physical limitation on how much AI we can deploy.
The irony is profound: we’re building systems that are a trillion times less energy-efficient than the human brain to augment - and potentially replace - human labor. Yet the human brain itself represents the gold standard of computational efficiency that we’re nowhere close to matching.
This creates several interesting dynamics:
The optimization imperative: As energy costs remain high and chip prices fall, the competitive advantage shifts to those who can extract maximum value from each watt consumed. Companies that can achieve better results with smaller, more efficient models will have a significant edge.
The infrastructure bottleneck: The 29 GW of power required for AI data centers by 2026 is equivalent to the output of roughly 30 nuclear power plants. This isn’t just a cost issue, it’s a question of whether the physical infrastructure can even support the AI buildout that companies are planning.
The economic crossover point: Despite the massive energy consumption, AI is already approaching and in some cases crossing the point where it’s economically advantageous to use it instead of human labor for specific tasks. Training a model costs millions, but deploying it at scale costs mere electricity and hardware.
The strategic race: Bostrom’s scenario of a first-mover advantage in superintelligence becomes even more critical when viewed through the lens of energy economics. The first organization to create a truly self-improving AI won’t just have a technological advantage, they’ll have an economic moat measured in orders of magnitude of efficiency improvement.
The Real Meaning for the Future
We’re living through a strange transitional period. AI systems are simultaneously:
Vastly less efficient than human brains in energy terms Increasingly more economical than human labor in specific domains
Rapidly improving in both capability and efficiency
Fundamentally reshaping how work gets done
Here is the real revolution: AI is unlocking human potential that was always there but constrained by the mechanics of execution. When a business analyst can prototype a working application overnight, when a marketer can test a hundred variations of a campaign in an afternoon, when a consultant can analyze patterns across thousands of projects in minutes - we’re removing the friction between insight and implementation.
The energy economics are daunting, but they’re also temporary. Moore’s Law may be slowing for traditional computing, but we’re in the early exponential phase of AI efficiency improvements. Each generation of models becomes more capable while requiring less compute. Each generation of chips delivers more operations per watt.
The brain’s 10¹² times better energy efficiency is a proof that radically more efficient intelligence is physically possible.
For now, we’re in a race: between the exponential growth of AI capabilities and adoption, and the linear constraints of energy infrastructure and human adaptation. The organizations and individuals who understand this dynamic and position themselves accordingly will thrive. Those who wait for the transition to complete will find themselves obsolete.
The energy costs are enormous, but the opportunity costs of not adapting are even greater.