How Is Anthropic Making So Much Money?


Almost everyone knows ChatGPT.

Claude is less famous.

Yet something surprising happened in the second quarter of 2026.

Anthropic, the company behind Claude, generated about $11.6 billion in quarterly revenue.

OpenAI generated about $6.7 billion.

For that quarter, the less famous AI company made roughly 1.7 times as much revenue.

And that may not even be the most surprising number.

Anthropic's annualized revenue run rate — roughly what its current pace would look like over a full year — went from:

About $9 billion at the end of 2025

to

More than $47 billion in May 2026

to

More than $65 billion by the end of July.

That's more than seven times the year-end 2025 level.

Now Anthropic has confidentially filed for a U.S. IPO, and a valuation around $2 trillion has been discussed.

So there is a much more interesting question than how popular Claude is.

Who is paying Anthropic all this money?

The Answer Isn't Just Chatbots

When most people think about Claude, they imagine a chatbot.

Type a question.

Get an answer.

Pay a monthly subscription.

But Anthropic's business is increasingly about something much larger:

Companies paying AI to do actual work.

One of the clearest examples is software development.

Anthropic launched Claude Code publicly in 2025.

It allows developers to ask Claude to write code, find bugs, modify software and perform other programming tasks.

Six months after becoming generally available, Claude Code reached $1 billion in run-rate revenue.

By February 2026, that number had climbed above:

$2.5 billion.

Anthropic said Claude Code's run-rate revenue had more than doubled since the beginning of the year.

That's an important clue to where AI money may be going.

AI Can Be Much Bigger When Companies Pay for It

An individual might pay a few tens of dollars a month for an AI subscription.

A company can be very different.

It might have:

10 employees using Claude.

Then 100.

Then 1,000.

And companies can connect Claude directly to their own software through APIs.

One team might begin using it for coding.

Then another team adopts it.

Then customer service.

Data analysis.

Internal research.

Operations.

The AI stops being something employees occasionally chat with.

It starts becoming part of how the company works.

That can produce a very different business from selling individual chatbot subscriptions.

Then Where Does the $2 Trillion Number Come From?

Anthropic confidentially filed for a U.S. IPO in June 2026.

Shortly before that, it raised $65 billion at a $965 billion post-money valuation.

Now valuations around $2 trillion are being discussed for its public-market debut.

Two trillion dollars sounds extraordinary.

So let's do a very simple calculation.

Anthropic's July annualized revenue run rate:

$65B+

Possible valuation:

$2,000B

Divide one by the other:

$2,000B ÷ $65B ≈ 31

That's roughly 31 times its current annualized revenue pace.

But the people thinking about a $2 trillion Anthropic aren't only looking at today's business.

Reuters reported that Anthropic is forecasting $190 billion to $200 billion in revenue in 2028.

Now do the calculation again:

$2,000B ÷ $200B = 10

Suddenly the number looks very different.

That tells us something important.

A $2 trillion Anthropic isn't really a story about what the company is today.

It's a story about what people believe it could become.

And that creates another question.

Can Anthropic Really Reach $200 Billion?

Going from a $65 billion annualized revenue pace to around $200 billion would still require enormous growth.

More companies would need to use Claude.

Existing customers would need to give Claude more work.

Products such as Claude Code would need to keep expanding.

And Anthropic isn't alone.

OpenAI is competing for many of the same customers.

So are Google and Meta.

Cheaper models are appearing around the world.

Being the smartest AI may not be enough.

And we're already seeing evidence of that.

The Best AI Doesn't Always Win

Anthropic recently released its most advanced model, Fable 5.

You might expect companies to immediately move to the best model available.

They didn't.

Financial Times reported that Fable 5 accounted for only about 11% of U.S. customer spending on Anthropic models, with many customers choosing cheaper options instead.

Why?

Price matters.

Imagine one AI model is 10% better but costs twice as much.

Does every task really need it?

Probably not.

A company might use:

a cheaper model for simple work,

a mid-range model for coding,

and the most powerful model only for difficult problems.

That changes the AI race.

The question is no longer only:

Who has the smartest AI?

Increasingly, it's also:

Who offers enough intelligence at the right price?

But Revenue Isn't the Same as Money Kept

This may be the most important part of Anthropic's business.

Traditional software can have wonderful economics.

Build the software once.

Sell another copy.

The additional cost can be relatively small.

AI works differently.

Every time Claude does work, computers are working too.

Those computers require chips.

Data centers.

Electricity.

Networking.

Cloud infrastructure.

So when Anthropic generates another dollar of revenue, part of that dollar must pay for the computing required to deliver Claude.

This means that a rapidly growing AI company can have two things happening at the same time:

More usage creates more revenue.

But:

More usage also creates more cost.

So how much is Anthropic actually keeping?

What Happens to $100 of Claude Revenue?

Anthropic doesn't publicly disclose exactly how much the average customer generates or costs to serve.

But we have a useful clue.

Reuters Breakingviews, citing PitchBook estimates, recently put Anthropic's gross margin at roughly 44%.

Think about that in the simplest possible way.

For every:

$100 of revenue

roughly:

$56 goes toward direct costs

leaving around:

$44 of gross profit

before expenses such as research, employees, sales and other operating costs.

Now scale that up.

Suppose Anthropic eventually reaches:

$100 billion in revenue.

At that cost structure, direct costs would be around:

$56 billion.

Now imagine Anthropic finds a way to make its AI infrastructure just 10% more efficient.

Ten percent of $56 billion is:

$5.6 billion.

That's $5.6 billion potentially saved from a relatively modest efficiency improvement.

Reuters Breakingviews made essentially this calculation while examining why Anthropic has strong incentives to reduce its enormous compute bill.

And suddenly something becomes clear.

Making Claude cheaper to run can be worth billions of dollars.

This Changes What the AI Race Is About

For the past few years, most AI headlines have focused on intelligence.

Who has the best model?

Who scores highest on benchmarks?

Who can write better code?

Who can reason better?

Those things still matter.

But when AI becomes a business generating tens of billions of dollars, another competition becomes equally important.

Who can produce the same useful AI for less money?

A faster chip matters.

A cheaper model matters.

Better data-center utilization matters.

Lower electricity consumption matters.

Software that gets more work from the same computing hardware matters.

At Anthropic's scale, small improvements can move billions of dollars.

This also helps explain why the AI industry is spending so much money on chips, memory, data centers and infrastructure.

The infrastructure isn't sitting behind the business.

The infrastructure is part of the economics of the business.

Anthropic Is Starting to Show Another Important Number

There is an encouraging sign.

The Wall Street Journal reported that Anthropic produced a small operating profit in the second quarter of 2026, while generating its $11.6 billion of quarterly revenue.

That's significant.

Generating enormous revenue while spending enormous amounts on computing is one thing.

Showing that the business can begin producing operating profit is another.

But one profitable quarter doesn't settle the question.

Anthropic is still expanding extremely quickly.

Its infrastructure requirements are enormous.

Competition is increasing.

And customers are clearly becoming sensitive to AI pricing.

So the next phase may be harder than simply attracting more users.

Anthropic has to prove it can do two things at the same time:

Sell much more Claude.

And:

Make each dollar of Claude revenue more economical to produce.

Even the Banks Are Moving In

There is another sign of just how large Anthropic has become.

Ahead of its IPO, Anthropic is preparing a revolving credit facility that could exceed:

$10 billion.

Reuters reported that banks have been competing for positions in the facility.

Top-tier banks were reportedly asked for commitments of around $1.25 billion each.

Why would banks be so interested?

The lending business itself matters.

But there is something else nearby:

a potentially enormous IPO.

And that reveals how much economic activity is forming around companies like Anthropic.

Start with Claude.

Follow the money outward and you find:

AI models

↓

chips

↓

data centers

↓

electricity

↓

cloud infrastructure

↓

banks

↓

public markets

A simple Claude window sits on top of an enormous industrial system.

So Is Anthropic Really Worth $2 Trillion?

Nobody knows yet.

But we can understand why the number exists.

Anthropic isn't simply selling access to a chatbot.

Companies are putting Claude into real workflows.

Claude Code has become a multibillion-dollar business at an extraordinary speed.

Anthropic's annualized revenue pace went from roughly $9 billion at the end of 2025 to more than $65 billion by July 2026.

And the company believes revenue could eventually approach $200 billion in 2028.

But there is another side to the story.

Customers aren't automatically choosing the most expensive AI.

Computing Claude costs enormous amounts of money.

And competitors are pushing prices down.

So perhaps the most revealing Anthropic number isn't:

How many people use Claude?

Or even:

Is Anthropic worth $2 trillion?

It may be this:

When Claude generates $100 of revenue, how much can Anthropic eventually keep?

Today, estimates suggest roughly $44 remains as gross profit before the rest of the company is paid for.

If Anthropic can turn that $44 into $50, $60 or more while revenue keeps growing, the economics become dramatically stronger.

If it can't, enormous revenue alone may not tell the whole story.

That's why the next AI race may not simply be about building the smartest machine.

It may also be about something much less glamorous:

Who can make intelligence cheapest to produce?

And at this scale, even a small answer can be worth billions.

BEYOND THE OBVIOUS.

Sources