AI's Free Lunch is Ending

The second phase of the artificial intelligence transformation will be shaped by the economy, not technology.

For the past few years, when discussing AI in companies, we've always revolved around the same questions: Which model should we use? ChatGPT or Claude? Where should Copilot come into play? Which processes can we automate? How do we provide prompt training to employees? Which use case should we start with?

These were all valid questions. Because the early stages of artificial intelligence were essentially about capability , about discovering what the technology could do.

But in my opinion, we are now entering a new era.

The question of this era will not be "What can AI do?".

We'll ask a more difficult question:

Does what AI does create economic value?

And that's precisely why I believe the second phase of the AI transformation will be shaped not by technology, but by the economy .

An article published in Harvard Business Review in August 2026, titled "How to Respond to the Coming AI Cost Shock," provides one of the strongest signals of this change. The phrase used in the article is quite striking:

“The corporate AI era is on the verge of a massive shift.”

The era of enterprise AI is said to be on the verge of a major transformation.

This isn't because technology is getting worse. On the contrary, AI is becoming increasingly better, faster, and more capable.

The problem is different: We are starting to use AI more and more.

Until now, we may not have fully felt the true cost of enterprise AI adoption. Software companies have borne a significant portion of the costs of GPUs, inference, and tokens to acquire customers, establish usage habits, and integrate their products into organizational workflows.

HBR describes it in a much more provocative way:

“Organizations are effectively being paid to adopt AI.”

In other words, companies are, in a sense, encouraged to use AI, and even part of the initial cost is subsidized by technology providers.

This is not a model we are unfamiliar with in the world of technology.

First, you increase usage. You create a habit. You integrate the product into daily life. Then the economic model begins to mature.

In AI, we may now be approaching exactly that stage.

Because artificial intelligence is not like classic enterprise software.

Calculating a software license is relatively simple. You have a thousand employees, you pay a monthly license fee per employee, and you know your approximate cost.

However, the calculation doesn't work that way with AI agents.

How many times a day will an agent run? How much data will it read? How many files will it analyze? Which model will it connect to? How many alternatives will it generate? Will it trigger another agent? Will it perform searches? Will it execute code? Will it perform an operation on a system?

Each of these consumes computing power.

Therefore, the cost model is gradually evolving into this:

User → Task → Token → Inference → Compute → Action → Cost

And this is where the issue moves from the technology department to the CFO's desk.

Because AI is no longer just a technology investment.

A business economics.

When we combine this microeconomic picture with the Reuters news report dated August 21, 2026, the picture becomes even more striking.

According to Reuters, citing BNP Paribas data, AI hyperscalers have issued approximately $220 billion worth of debt by August 10, 2026.

At the same time a year ago?

$12.5 billion.

The difference is approximately $207 billion.

This figure tells us something very important in the AI story.

Artificial intelligence appears to us as a small window on the screen. We type a question, and a few seconds later the answer arrives. The experience is highly digital and almost abstract.

But behind that answer lies a very physical world.

There are chips.

There are servers.

There are data centers.

There is electricity.

They have cooling systems.

Fiber optic connections are available.

There is land available.

There is energy infrastructure.

And behind all of this is a massive amount of capital .

Considering that hyperscalers like Microsoft, Amazon, Alphabet, and Meta are expected to reach a total capital expenditure of approximately $725 billion in 2026, it's clear that AI is no longer just a software revolution.

AI is transforming into one of the biggest physical infrastructure investment cycles the world has ever seen.

Therefore, from now on, it will not be enough for executives who want to understand AI to simply follow the new version of GPT.

They will need to understand Compute.

They will need to understand energy.

They will need to understand CapEx.

They will need to understand the cost of borrowing.

They will need to understand free cash flow.

Because these will be the languages of the second era of AI.

There is a paradox here that I find particularly important.

The unit cost of AI could decrease.

The cost of one million tokens could become much lower than it is today. Models could operate more efficiently. Chips could be more powerful.

But the total amount of artificial intelligence we use simultaneously could increase much faster.

When an employee uses AI a few times a day, the cost may seem insignificant.

But in a world where hundreds of AI agents work around the clock, researching, analyzing customers, preparing proposals, monitoring systems, making decisions, and managing other agents on behalf of the company, we're no longer talking about the same old calculations.

This reminds us of an old economic principle: when something becomes cheaper, total spending doesn't always fall. Sometimes usage grows so rapidly that total cost, on the contrary, rises.

We might experience a similar situation with AI.

Therefore, I believe companies need to learn a new concept:

AI Unit Economics.

It's no longer enough for an AI application to simply be working.

He needs to work to be economically stable.

For example, imagine you have an AI sales agent.

That agent might be researching a hundred potential clients a day.

It's very impressive at first glance.

But the real questions are different:

What is the AI cost of this research?

How many hours does this save the sales employee?

Does it increase the conversion rate?

Does it shorten the sales cycle?

Does it reduce the time spent on the wrong potential customers?

Does it generate extra income?

Therefore, we need to evaluate the use of AI not only by its activity but also by its results.

I simply think of it this way:

Increased revenue + time savings + improved quality + reduced risk.

minus

AI usage cost + infrastructure + integration + governance cost

equals

Net AI Value.

This perspective is important because, in my opinion, one of the most dangerous KPIs that companies use today is AI adoption rate.

"75% of our employees use AI."

Beautiful.

But this tells us almost nothing about its economic value.

If a factory manager came to us and said, "We increased our electricity consumption by 40 percent this month," we wouldn't applaud that as an achievement.

Our first question would be:

What did you produce in return?

We need the same disciplined thinking for AI as well.

Usage may have increased.

What about income?

What about the cost?

What about customer value?

What about decision quality?

What about people shifting their time to more valuable tasks?

I think in the future, companies will start talking about metrics like AI Value per Dollar .

A more strategic equivalent of this could be called AI Capital Productivity™ :

For every unit of capital we invest in AI, how much economic value do we generate?

This question leads us to another very important point.

Up until now, in the AI transformation, we have often asked the following question:

"Can AI do this?"

I think the question for the second term will be different:

"Is it economically worthwhile for AI to do this?"

There is a huge difference between these two questions.

Capability and value are not the same thing.

Not everything that is technologically feasible is economically worthwhile.

A job might be done by AI, but humans could still be cheaper.

AI might be faster, but the customer might want human interaction at that point.

AI can perform an analysis perfectly, but the risk cost of making the wrong decision can be high.

Or, conversely, AI could create such a significant advantage in speed, quality, and scale that it would become economically impossible for a human to do it alone.

Therefore, in my opinion, the question of "Human or AI?" will be too simplistic for the organizations of the future.

The right question is:

Which combination of intelligence creates the highest overall value in which project?

Some tasks will remain entirely in human hands.

Some will switch entirely to AI.

But perhaps the greatest value will be created in areas where humans and AI work together.

At this point, it's necessary to add an economic dimension to the Delegation Engineering™ concept I used earlier.

We need to consider not only which tasks we will delegate to AI, but also whether delegating those tasks to AI makes economic sense.

So the future leader will not only allocate human resources and capital.

It will also allocate intelligence .

How much attention does each problem receive?

How much artificial intelligence capacity does it have?

How much capital?

How much decision-making authority?

I call it Intelligence Allocation™ .

Because in the age of AI, perhaps one of the most important tasks of a manager will be to allocate human intelligence and artificial intelligence to the right problem and within the right economic model.

There is also an important lesson we can learn from the internet era of the 2000s.

Has the internet changed the world?

Definitely.

But were all investments made in the internet age good investments?

Absolutely not.

Both sentences can be true at the same time.

The same applies to AI.

Artificial intelligence could radically change the world.

And also:

Some AI investments can be extremely bad economically.

Just because a technology is transformative doesn't mean every investment in it is the right one.

This is where the role of strategy becomes important again.

Today, Alibaba raising approximately $10 billion in new capital to finance its AI investments shows that this story is not unique to the US.

The need for capital in AI infrastructure is becoming global.

Companies are sacrificing today's profits to invest in the AI capabilities of the future.

This could be perfectly rational.

But in the end, everyone has to come back to the same question:

When and how will this investment generate economic value?

I think the CFO's voice will be increasingly heard in AI discussions over the next few years.

And that's not a bad thing.

On the contrary, it's a sign that the AI transformation is maturing.

Initially, technology teams tried it.

In the second phase, business units will take ownership.

The CFO will bring economic discipline.

CHRO will redesign the division of labor between humans and AI.

It will change COO workflows.

The CEO will then decide which strategic future all of this serves.

The questions the board of directors needs to ask will also change.

Instead of asking, "How many AI projects do we have?":

Which of our AI investments are truly creating value?

Instead of asking, "Which model are we using?", ask:

What is the return on this investment?

Instead of asking, "Do our employees use AI?", ask:

Has AI increased our organization's economic capacity?

And perhaps the most crucial question:

What were the alternative uses for every dollar we spent on AI?

Because all strategies ultimately become a reality through resource allocation.

I am optimistic about the future of AI.

But believing in AI shouldn't mean not questioning the AI economy.

On the contrary.

As AI becomes more strategic, economic discipline must become all the more important.

Because now we're not just talking about a few software licenses.

Data centers worth hundreds of billions of dollars.

Energy infrastructure.

Chips.

Borrowing.

Cost of capital.

And increasingly autonomous AI agents.

At the end of this whole chain, only one question awaits us:

What is the economic value produced?

In the early stages of the artificial intelligence transformation, the big question was:

"What can AI do?"

The question for the second term will be much more difficult:

"Does what AI does create economic value?"

I believe that the next AI race will be won not by companies that use technology the most, but by those that can transform technology into economic value in the most disciplined way .

Because using AI will become increasingly easier.

However, managing AI economically will become increasingly difficult.

And precisely for this reason:

The second phase of the AI transformation will be shaped by the economy, not technology.