The Inteligence Economy

In 2025, OpenAI reportedly generated $13.07 billion in revenue but recorded a net loss of $38.5 billion. Its operating loss was $20.92 billion. In the first half of 2026, Anthropic reportedly recorded a net loss of $80.7 billion, according to figures from a draft prospectus for its planned initial public offering.

At first glance, these numbers seem absurd. Two of the companies at the center of the artificial intelligence revolution are generating billions in revenue while reporting losses that dwarf the earnings of many established businesses.

But there's an important distinction: Anthropic's reported loss included approximately $72 billion in non-cash accounting charges related to the valuation of financial instruments that could convert into shares. That is not the same as spending $80.7 billion to operate its business. OpenAI's reported 2025 net loss was also significantly affected by accounting charges associated with its financing structure.

The headlines tell one story. The underlying financial details tell another.

And that raises a much more interesting question.

Where is all the money in artificial intelligence actually going?

The race is about more than intelligence

For the past few years, the AI industry has been moving at a remarkable pace.

OpenAI has ChatGPT. Google has Gemini. Anthropic has Claude. Meta is investing in its own AI models, while Microsoft, Amazon, and other technology companies are investing heavily in the infrastructure that makes these systems possible.

Every major player seems to be racing toward the same destination: more capable AI, more users, more computing power, and a larger share of what could become an enormous market.

But what does it actually mean to win?

Is it building the most intelligent model? Attracting the most users? Generating the most revenue? Becoming profitable? Or controlling the infrastructure on which everyone else depends?

These are different goals, and the company that wins one doesn't necessarily win the others.

A company might develop an exceptional AI model but struggle to cover the cost of running it. Another might develop a less capable model but build a more profitable business around it. A third might sell the chips and computing infrastructure that the entire industry relies on.

The AI race isn't just a competition between models. It's a competition between business models, infrastructure providers, investors, and companies trying to determine where the economic value of artificial intelligence will ultimately accumulate.

To understand who might win, we need to understand what everyone is paying for.

Where does the money actually go?

Building an AI product is not simply a matter of training a model and making it available to users.

Behind every response generated by an AI assistant is an infrastructure layer that can include specialized chips, servers, data centers, networking equipment, electricity, and cooling systems.

There is also the cost of training models, hiring researchers and engineers, conducting experiments, developing products, serving customers, and maintaining the software that makes the technology useful.

Consider the major categories of expenditure.

1. Computing power

AI models require enormous amounts of computation, both during training and when generating responses for users.

That computation runs on specialized hardware, including GPUs and other AI accelerators. Companies must purchase or rent access to this hardware, often through cloud providers.

The costs don't stop when a model is released.

Every question asked, image generated, document analyzed, or piece of code written consumes computing resources. As usage grows, so can the cost of providing the service.

A company can attract millions of users and generate substantial revenue while still facing difficult questions about how much it costs to serve those users.

The challenge is to make the service valuable enough that customers are willing to pay more than it costs to provide.

2. Data centers and energy

The chips need somewhere to operate.

That means data centers, power infrastructure, cooling systems, networking equipment, and the facilities required to keep everything running.

These investments can involve enormous sums of money and long-term commitments.

They also create an important distinction between spending money today and recovering that investment over time.

A data center may support AI workloads for years, but the economics depend on how much it costs to build and operate, how heavily it is used, and how much revenue the infrastructure helps generate.

If demand grows as expected, the investment may prove valuable.

If demand disappoints, computing prices fall faster than expected, or newer hardware makes existing equipment less competitive, recovering the cost becomes more difficult.

The question isn't simply whether AI needs more infrastructure.

It's whether the demand for that infrastructure will justify the investment.

3. Research and development

AI companies also spend heavily on research.

They hire researchers and engineers, train models, run experiments, improve reliability, and develop new capabilities.

These investments are intended to produce better products and create advantages over competitors.

But research doesn't guarantee commercial success.

A breakthrough may produce a more capable model without generating enough additional revenue to justify its cost. Competitors may replicate the improvement. Customers may also decide that an existing, cheaper model is good enough.

In a competitive market, better technology can be essential to survival without necessarily translating into better profit margins.

That creates a difficult balancing act: companies need to keep improving their products while finding ways to make those products economically sustainable.

4. Everything around the model

There are also expenses that are less visible to ordinary users.

Companies need sales teams, customer support, security systems, legal services, marketing, and the software infrastructure required to deliver products reliably.

Enterprise customers may require additional security, compliance, and administrative features. AI companies also need to manage the legal and operational risks associated with their products.

None of these expenses is unique to AI, but they add to the cost of turning a powerful technology into a functioning business.

The result is an industry in which developing an impressive model is only part of the challenge.

The other part is building a business that can afford to operate it.

If the companies are losing money, why do investors keep funding them?

This is where the story becomes more interesting.

Investors don't necessarily expect a company to be profitable immediately. They may be willing to fund years of losses if they believe the business can eventually generate returns large enough to justify the risk.

The logic behind investing in AI is that the technology could become a foundational part of the global economy.

If AI becomes essential to software development, customer service, scientific research, business operations, and countless other activities, the companies that control important parts of the technology could become extraordinarily valuable.

Investors want exposure to that possibility.

But there is another factor: competition.

Imagine that you are an investor considering an AI company.

You believe the technology could transform entire industries. You also know that competing companies are investing billions of dollars in research, infrastructure, and talent.

If you refuse to invest, a competitor might gain an advantage that becomes difficult to overcome.

You might therefore decide that the risk of investing is preferable to the risk of missing out entirely.

Now imagine thousands of investors, executives, and companies making similar decisions.

The result can be a race in which everyone feels pressure to spend more because everyone else is spending more.

That doesn't automatically mean the investments are irrational. It does mean that competitive pressure can influence how much money companies are willing to commit before the long-term returns become clear.

And that leads to another question.

What happens when investors start demanding returns instead of promises?

Is investor money really unlimited?

It can sometimes look that way.

AI companies announce enormous funding rounds. Technology giants commit billions to infrastructure. Investors compete for stakes in businesses they believe could shape the next era of computing.

But investor money is not unlimited.

Venture capital firms raise money from their own investors. Those investors have financial obligations and expectations. Public companies allocate capital while balancing competing priorities. Banks and other lenders assess whether borrowers can repay their debts.

Every source of capital has limits, even when those limits aren't immediately visible.

The important question is how long investors are willing to finance a company before its business model demonstrates that it can generate sufficient returns.

A company might continue operating at a loss because investors believe its future prospects justify additional funding. It might also have access to substantial cash reserves, strategic partnerships, or revenue from other parts of its business.

But access to capital can change.

Investors may become more cautious. Financing may become more expensive. Expectations may shift. A company that previously had little difficulty raising money might discover that its next funding round comes with less favorable terms.

None of this means that the AI industry is about to run out of money.

It means that the industry's ability to spend should not be confused with an unlimited ability to finance losses indefinitely.

Eventually, investors need a credible explanation of how the money they commit will produce a return.

The fear of missing out

The AI boom also reveals something about human behavior.

When investors see a technology attracting enormous attention, they may worry that waiting will leave them behind.

Companies experience a similar pressure.

If competitors are building AI capabilities, executives may fear that delaying investment will leave their businesses at a disadvantage. Even companies that are uncertain about the returns may feel compelled to participate.

Governments face a related concern: if AI becomes strategically important, falling behind could have economic, scientific, or national security consequences.

These incentives can reinforce one another.

Investors fund companies because they expect AI to become important. Companies invest to avoid falling behind. Infrastructure providers expand to meet expected demand. Governments support research and domestic capabilities because they don't want other countries to gain an overwhelming advantage.

As more money enters the industry, the scale of the opportunity appears even larger, attracting further investment.

But expectations can become disconnected from outcomes.

A company can invest based on the assumption that demand will continue growing rapidly. An infrastructure provider can expand based on forecasts of future computing requirements. An investor can value a business based on profits that may not materialize for years.

If those assumptions prove correct, the investments may generate substantial returns.

If they don't, some of the money invested may never be recovered.

The challenge is distinguishing genuine long-term demand from expectations that become more optimistic simply because everyone else is investing.

Who actually makes money from the AI boom?

One of the most interesting possibilities is that the biggest financial winners might not be the companies developing the most popular AI assistants.

Consider the different layers of the industry.

Chip manufacturers sell the hardware used to train and run AI models.

Cloud providers sell computing capacity and infrastructure.

Data center operators provide the facilities that house the equipment.

Energy companies supply the electricity needed to operate those facilities.

AI companies build models and products that customers use.

Software companies incorporate AI into products that already have paying customers.

Each layer has a different business model, cost structure, and relationship with the rest of the industry.

An AI company might spend billions purchasing computing capacity from a cloud provider. The cloud provider, in turn, might purchase hardware from a chip manufacturer and invest in the infrastructure required to support its customers.

Money moves through the ecosystem, but the economic benefits are not distributed equally.

Some companies may earn revenue from AI infrastructure investments long before the AI applications running on that infrastructure become profitable.

Others may struggle to monetize their technology even as demand for computing power increases.

This creates an important distinction between supplying the AI industry and building a successful business using AI.

The two are connected, but they are not the same.

It is possible for demand for AI computing to grow substantially while some AI application companies struggle financially.

It is also possible for infrastructure providers to overestimate demand and invest in capacity that takes longer than expected to generate adequate returns.

The question is not simply which company has the best technology.

It is which companies can turn their position in the AI ecosystem into durable economic value.

What happens if AI succeeds but its investors lose money?

This may be the most important question of all.

Imagine a future in which AI becomes extraordinarily useful.

Businesses use it to automate routine work, developers build software faster, researchers accelerate scientific discovery, and individuals gain access to capabilities that previously required specialized expertise.

Now imagine that the companies responsible for building this technology invested far more money than they eventually recovered.

Could AI transform the world while some of its investors lose money?

Yes.

Technological success and investment success are not identical.

The internet transformed commerce, communication, and entertainment, but the dot-com boom also produced companies whose valuations and business models proved unsustainable.

The technology survived. Many investments did not.

A similar outcome is possible in AI, although its eventual trajectory may be very different.

Competition could drive prices down. Open models could reduce the cost of accessing capable AI. Businesses could discover that customers are unwilling to pay as much as providers expected.

Alternatively, demand could grow enough to support substantial profits across multiple layers of the industry.

Both outcomes are possible. The evidence will emerge over time.

But one lesson from previous technological booms is worth remembering: a technology can be revolutionary without every company associated with it becoming a great investment.

The role of governments

The AI race is not exclusively a private-sector competition.

Governments have their own reasons for investing in AI research, supporting domestic infrastructure, regulating the industry, and securing access to advanced computing technology.

AI could affect economic productivity, military capabilities, cybersecurity, scientific research, public services, and international competitiveness.

That makes it more than a commercial opportunity.

Governments may support AI development even when the immediate financial returns are uncertain because they believe the technology has strategic importance.

They may also intervene to address risks involving privacy, competition, employment, intellectual property, and public safety.

This introduces another dimension to the economics of AI.

The industry's development will depend not only on what customers are willing to buy and what investors are willing to finance, but also on what governments permit, encourage, fund, or restrict.

The result is a complicated system of private investment, public policy, international competition, and technological development.

And the consequences will extend far beyond Silicon Valley.

So, who wins the AI race?

There is no definitive answer yet.

OpenAI, Google, Anthropic, Meta, Microsoft, and other companies occupy different positions in the industry. Their strengths, financial resources, distribution, infrastructure, and business models differ.

Some may develop more capable models. Others may become better at delivering AI cheaply and reliably. Some may dominate particular markets, while others may discover that their technology is easier to replicate than they expected.

The eventual winners may not even be the companies attracting the most attention today.

But judging the race solely by model capabilities, user numbers, funding rounds, or company valuations would miss the larger picture.

A company can raise billions without proving its business model. A company can generate enormous revenue without producing a profit. A company can lose money in the short term while making investments that eventually generate substantial returns.

Understanding the difference requires looking beyond the headlines.

It requires examining the financial statements, following the capital, understanding the infrastructure, and questioning the assumptions behind the industry's biggest investments.

Because the AI race is not just about who can build the smartest machine.

It is also about who can build a sustainable business around it, who controls the infrastructure, who captures the economic value, and who ultimately bears the risk.

The technology may change the world.

But that doesn't mean everyone funding the transformation will profit from it.

And until we understand where the money is going, we cannot fully understand who is winning.