One of the most appealing things about software companies was that growth did not necessarily require building a new factory for every customer.

A company writes the software once, then can sell it to thousands or millions of users at a relatively limited additional cost.

But AI is changing that equation again.

The more people use models, the more processors, data centers, power, storage, and networks a company needs. Worse still, some of this computing capacity must be reserved years before the actual level of demand is known.

This is where AI companies start to resemble infrastructure businesses more than traditional software companies.

What do Anthropic’s numbers reveal?

A filing Anthropic submitted in October 2026 revealed the enormous scale of its computing and infrastructure commitments.

Reuters reports that the company has committed to more than $600 billion in spending with several partners over roughly a decade, while The Register estimated the total value of the listed contracts—accounting for their differing terms and degrees of enforceability—at between $782 billion and $807 billion.

These relationships include companies such as Amazon, Google, Microsoft, Broadcom, and other computing providers.

What matters here is not just the number, but the nature of the commitment.

Some major contracts require payment even if Anthropic does not use all the available capacity, according to its filing. This means computing shifts from a cost that rises only with usage to a commitment that must be covered by future revenue.

What does reserving computing capacity in advance mean?

Think of it as an airline expecting a huge increase in passenger numbers.

To benefit from that growth, it cannot wait until the passengers arrive before it starts looking for planes.

It has to order the planes and secure airports and crew years in advance.

If demand materializes, these investments become the foundation for growth.

But if passenger numbers fall short of expectations, the planes and costs remain.

Anthropic faces a similar equation.

If it wants to train larger models and serve millions of customers, it must first ensure it has the computing capacity.

This is where operating leverage comes in

When a company has substantial fixed costs and commitments, its profits become highly sensitive to revenue levels.

If revenue growth outpaces those costs, profits can rise quickly.

But if demand slows, commitments do not fall at the same pace.

This is what is commonly known as operating leverage.

Anthropic is already growing rapidly: its revenue rose from $386 million in 2024 to $4.59 billion in 2025, then reached $16.29 billion in the first half of 2026, while its operating loss fell to $693 million over the same period.

But these figures must continue to grow enough to cover future commitments.

The company itself says that its ability to meet rising computing needs depends on achieving substantial revenue growth.

Why is the company taking this risk?

Because the potential payoff is enormous, too.

Anthropic says it believes that abundant AI capabilities could transform sectors of the global economy on a scale exceeding historic shifts such as electricity, semiconductors, and the internet.

If much of that comes to pass, companies with sufficient computing capacity may be able to sell products and services on a far greater scale than today.

So the company is not spending because computing is cheap, but because it fears that a shortage of computing itself could hold back its growth.

But computing does not guarantee revenue

And this is the most important economic lesson.

Owning more processors does not automatically mean earning more profit.

Computing capacity must lead to better models, then to products customers want to use, and finally to revenue that exceeds the cost of running that infrastructure.

If that does not happen as quickly as expected, the company may have to borrow or issue new shares, which could increase its debt or dilute the ownership of existing shareholders.

That is why the question in the AI race is no longer just:

Who has the most powerful model?

It is also:

Who can reserve computing capacity today, then find enough demand for it tomorrow?

The AI boom is making software more connected to the physical world than we are used to.

Behind every token are chips, electricity, data centers, and long-term contracts.

The winner will not necessarily be the one that spends the most, but the one that can turn these huge commitments into revenue and cash flow before its infrastructure shifts from a competitive advantage to a financial burden.