#AnthropicIPOFiling:HighGrowth,HighLosses

After going through the latest numbers from Anthropic’s IPO prospectus, the part that caught my attention is not simply the size of the company or what valuation the market may eventually give it.

It is the economics behind the AI race.

Anthropic generated approximately $4.59 billion in revenue during 2025, compared with $386 million a year earlier. That is roughly a twelvefold increase in one year. On the surface, that is an extraordinary growth story.

But then you look at what it took to generate that growth.

Anthropic reported an $8.06 billion operating loss, while compute and infrastructure expenses reached approximately $7.33 billion, around 58% of its operating expenses. The headline GAAP net loss was even larger at roughly $41.97 billion, although around $34 billion of that was related to a non-cash accounting charge connected to financing liabilities.

That distinction matters.

The $42 billion number is shocking, but it does not mean Anthropic literally spent $42 billion in cash that year. The operating loss and infrastructure spending give us a better picture of how expensive it is to build and operate increasingly powerful AI systems.

And this is where I think the bigger investment story starts.

AI companies are not simply spending money on software anymore.

They need enormous amounts of compute.

Compute needs chips.

Chips need memory and storage.

All of that hardware needs data centers.

Data centers need networking, cooling and electricity.

And once the infrastructure is built, somebody has to keep paying for the computing capacity required to train and serve the models.

That creates a very different economic structure from the traditional software boom.

The model companies are competing to build better intelligence and attract users. But underneath that competition sits an enormous infrastructure layer that has to expand regardless of which individual model ultimately captures the largest share of the market.

That is why I keep looking beyond the AI application layer.

The interesting question is not only “Which AI model wins?”

It is also:

“Who gets paid every time the AI industry needs more compute?”

The prospectus makes that question even more relevant.

Anthropic disclosed plans involving approximately $518 billion of future cloud, computing and infrastructure obligations, according to reporting on the prospectus. That is not the same thing as $518 billion of current spending, and the timing and structure of those commitments matter, but the scale shows how capital-intensive the next phase of AI development could become.

And this is not happening in isolation.

Anthropic has been expanding its relationships across the infrastructure stack, including agreements involving major cloud and computing providers. The company itself has previously said that new funding would be used to expand compute capacity, while infrastructure partners including Micron, Samsung and SK hynix are involved in supplying critical memory, storage and logic technologies.

That is the part I think the market will increasingly focus on.

The first phase of the AI trade was about the excitement around models.

The next phase may be much more about economics.

How much does each additional user cost?

How much compute is required to serve increasingly sophisticated models?

How quickly does revenue grow compared with infrastructure spending?

How much capital has to be committed before an AI company can reach sustainable free cash flow?

Those questions become more important as the industry gets larger.

And there is another detail worth watching: Anthropic's revenue is growing extremely quickly, but customer concentration also matters. Reporting on the prospectus indicates that the company's two largest direct customers each represented about 12% of 2025 revenue, meaning almost a quarter came from those two customers combined.

So the story is not simply “AI is growing.”

The more useful story is:

AI is growing, but the infrastructure bill is growing with it.

That is why I don't think investors should automatically assume every company associated with AI will benefit in the same way.

Model companies have to compete for customers.

Infrastructure companies can potentially earn revenue from multiple parts of the AI ecosystem.

The companies providing the underlying compute, memory, storage, networking, data-center capacity and power are effectively supplying the machinery required for the entire race.

This does not make infrastructure companies risk-free. Their businesses also face enormous capital requirements, supply constraints, customer concentration and the possibility that AI spending eventually slows.

But Anthropic's numbers give us a fresh look at just how much physical infrastructure sits underneath the seemingly simple experience of opening an AI application and asking a question.

The AI boom may look digital from the outside.

The bill is extremely physical.

Chips.

Memory.

Storage.

Servers.

Data centers.

Networks.

Electricity.

Cooling.

Cloud capacity.

All of it has to be paid for.

That is why I am paying more attention to the AI infrastructure economy instead of chasing every new AI concept that appears in the market.

The model race is still important.

But the companies building and supplying the infrastructure behind that race may become an equally important part of the story.

Anthropic's prospectus doesn't prove which companies will ultimately create the most shareholder value.

What it does show is something much more concrete:

Building frontier AI at scale requires extraordinary amounts of capital and infrastructure.

Revenue can grow 1,088% in a year.

But if the cost of compute and infrastructure is growing alongside it, the next question for investors becomes much harder:

When does AI growth turn into sustainable cash generation?

That is the question I will be watching in the next phase of this AI cycle.

Not just who has the smartest model.

Not just who has the most users.

But who can convert AI demand into durable cash flow while controlling the enormous infrastructure bill underneath it.

The AI story is clearly still developing.

But the economics are becoming impossible to ignore.

$NVDA