Anthropic Opens the Books: The AI Bubble Argument Just Got More Complicated
Anthropic's IPO filing makes it much harder to argue that AI demand is fake. It does not make it harder to argue that investors may be spending too much money to satisfy it.
By Jeff FergusonOctober 28, 20269 minute read
For much of the AI boom, one of the hardest questions to answer has been remarkably basic: Is anyone actually making money from this?
We could see the spending. Nvidia could tell us how many chips it was selling. Amazon, Microsoft, Google and Meta could tell us how much they were investing in data centers. Utilities could tell us how much electricity was being requested. Banks could tell us how much financing was being arranged.
But the economics of the companies building the frontier AI models were largely private. OpenAI is private. Anthropic was private. Revenue numbers arrived through leaks, investor presentations and people familiar with the matter. We had useful indications. We did not have the kind of financial disclosure investors normally use to evaluate a company.
Anthropic's IPO filing changes that. And it complicates the bubble story.
The Revenue Is Real
Anthropic generated $41 million of revenue in 2023, then $386 million in 2024, then $4.585 billion in 2025. During the first six months of 2026 alone, revenue reached $16.286 billion. The S-1 turns what had been estimates into a public financial record.
You do not have to like AI valuations to recognize what those numbers mean. This is not a business waiting for somebody to discover a reason to buy its product. Customers already are.
By September, Anthropic had more than 11,900 customers generating at least $100,000 in trailing-twelve-month revenue, including roughly 250 above $10 million. That doesn't prove every customer is receiving a positive return. It does prove there is real demand at meaningful scale.
The Headline Loss Is Real — But Misleading
The number most likely to create dramatic headlines is Anthropic's enormous first-half net loss. But most of that loss reflects changes in the value of convertible financing and related liabilities as Anthropic's valuation increased.
The operating picture is very different: Anthropic reported a $693 million GAAP operating loss for the first half, along with $1.618 billion of adjusted operating income and roughly $3.9 billion of operating cash flow. In the second quarter alone, it reported $11.6 billion of revenue and $1.0 billion of GAAP operating income.
A company generating $16.3 billion in six-month revenue while approaching operating profitability is not the same economic story as a company burning cash while searching for customers.
It Also Has an Astonishing Amount of Liquidity
The filing shows Anthropic with a very large liquidity position after raising extraordinary amounts of private capital. That matters for how we think about financing risk.
Anthropic is not a fragile startup with a few months of runway trying to refinance a single data-center loan. Its balance sheet gives it room to absorb volatility.
Then the filing reveals the other side.
The Commitments Are Also Astonishing
Anthropic needs enormous amounts of compute. The S-1 reports $8.945 billion of compute and infrastructure expense in the first half of 2026 and describes long-term purchase commitments, take-or-pay arrangements and infrastructure relationships extending for years.
Reuters reports compute commitments exceeding $600 billion. The filing itself makes clear that the improved economics still depend on an extraordinary future spending path.
So Anthropic simultaneously has enormous revenue growth, enormous liquidity and enormous future obligations. Those facts belong in the same conversation.
This Is Why “Bubble” Is Such an Unhelpful Binary
People tend to talk about bubbles as though the underlying thing must be fake. It isn't.
The internet bubble did not prove that the internet was useless. Railway booms did not prove that railroads had no economic value. A technology can transform the economy and still attract more capital than the eventual returns justify.
That may be the most important lesson in Anthropic's prospectus. Claude can be extremely useful. Anthropic can become enormously profitable. AI can reshape entire industries. And investors can still overpay for the companies and infrastructure built around it. All four things can be true.
Amazon, Google and the Rest of the Stack Are Everywhere in This Story
The filing also exposes just how intertwined the AI economy has become. Anthropic depends heavily on a limited group of cloud, hardware and distribution partners. Those firms can simultaneously be investors, suppliers, distributors, customers and competitors.
There is nothing inherently improper about that. It may be an extremely efficient business arrangement. But it makes the AI economy difficult to analyze using simple categories like customer, supplier, investor and competitor.
The same companies occupy several categories at once. That matters when trying to understand where losses would land if the cycle turns.
Customer Concentration Is Better Than You Might Expect — But Still Meaningful
Anthropic has an enormous customer base. That reduces some risk. But concentration remains material, and many customers are not locked into contracts for the full life of the infrastructure being built to serve them.
That creates an interesting mismatch. Infrastructure can be committed for many years. Customer demand can change much faster. The business works beautifully if usage keeps expanding. It becomes more difficult if compute commitments are fixed while customers remain flexible.
The Filing Changes My View — Just Not in the Obvious Way
Before the IPO filing, Anthropic was one of the larger unknowns in the AI investment cycle. We knew it was growing. We knew it was raising enormous amounts of money. We knew it had major partnerships. We did not know enough about the operating business to confidently say how much of that infrastructure spending was being supported by actual economic activity.
Now we know more. The revenue is substantially stronger than the private-company opacity suggested. The operating trajectory is much better than the headline net-loss number implies. The balance sheet is enormous.
That makes me less concerned about whether there is real demand for frontier AI. It does not make me less concerned about the scale of the investment cycle surrounding it.
The Bigger Question Is No Longer “Does Anyone Want AI?”
They do.
The more useful question is: How much capital should rationally be invested to meet that demand?
A business can be great. Its stock can still be overpriced. A product can be transformative. Its infrastructure can still be overbuilt. Revenue can grow rapidly. Financing commitments can still assume even faster future growth.
This is why EconRisk separates Demand / ROI from CapEx Sustainability, Financing / Credit and Market Concentration. Anthropic's filing lowers one kind of risk while increasing our confidence in several others.
The IPO Will Become a Price Discovery Event for the Entire AI Economy
Anthropic is applying to trade on Nasdaq under ANT. The final price has not yet been set. That may ultimately be more important than any single number in the prospectus.
Private investors have been valuing leading AI companies using private financing rounds, strategic investments and negotiated transactions. The IPO will let the public market decide what one of those businesses is worth every day.
That gives the AI boom something it has lacked: continuous public price discovery for a frontier-model company.
If investors accept an enormous valuation, that will strengthen confidence in the sector. If they demand a major discount, it will tell us something equally important. Either way, Anthropic becomes a benchmark.
The AI Bubble Argument Just Got Better
Not because Anthropic looks terrible. Quite the opposite.
A credible bubble argument should be able to survive evidence that the underlying technology works.
Anthropic makes that possible. The company is growing spectacularly. The demand is real. The customers are real. The revenue is real. And the economic potential may be enormous.
The remaining question is whether the amount of money being committed around that opportunity is rational relative to the returns it will eventually produce.
That is a much harder question than asking whether AI is useful. It is also the question that matters.
Commentary reflects the author's interpretation of the evidence. AI assisted the research and drafting. Sources are linked throughout. Read our AI disclosure.
Commentary · AI Economics & Valuation
Anthropic Opens the Books: The AI Bubble Argument Just Got More Complicated
Anthropic's IPO filing makes it much harder to argue that AI demand is fake. It does not make it harder to argue that investors may be spending too much money to satisfy it.
For much of the AI boom, one of the hardest questions to answer has been remarkably basic: Is anyone actually making money from this?
We could see the spending. Nvidia could tell us how many chips it was selling. Amazon, Microsoft, Google and Meta could tell us how much they were investing in data centers. Utilities could tell us how much electricity was being requested. Banks could tell us how much financing was being arranged.
But the economics of the companies building the frontier AI models were largely private. OpenAI is private. Anthropic was private. Revenue numbers arrived through leaks, investor presentations and people familiar with the matter. We had useful indications. We did not have the kind of financial disclosure investors normally use to evaluate a company.
Anthropic's IPO filing changes that. And it complicates the bubble story.
The Revenue Is Real
Anthropic generated $41 million of revenue in 2023, then $386 million in 2024, then $4.585 billion in 2025. During the first six months of 2026 alone, revenue reached $16.286 billion. The S-1 turns what had been estimates into a public financial record.
You do not have to like AI valuations to recognize what those numbers mean. This is not a business waiting for somebody to discover a reason to buy its product. Customers already are.
By September, Anthropic had more than 11,900 customers generating at least $100,000 in trailing-twelve-month revenue, including roughly 250 above $10 million. That doesn't prove every customer is receiving a positive return. It does prove there is real demand at meaningful scale.
The Headline Loss Is Real — But Misleading
The number most likely to create dramatic headlines is Anthropic's enormous first-half net loss. But most of that loss reflects changes in the value of convertible financing and related liabilities as Anthropic's valuation increased.
The operating picture is very different: Anthropic reported a $693 million GAAP operating loss for the first half, along with $1.618 billion of adjusted operating income and roughly $3.9 billion of operating cash flow. In the second quarter alone, it reported $11.6 billion of revenue and $1.0 billion of GAAP operating income.
A company generating $16.3 billion in six-month revenue while approaching operating profitability is not the same economic story as a company burning cash while searching for customers.
It Also Has an Astonishing Amount of Liquidity
The filing shows Anthropic with a very large liquidity position after raising extraordinary amounts of private capital. That matters for how we think about financing risk.
Anthropic is not a fragile startup with a few months of runway trying to refinance a single data-center loan. Its balance sheet gives it room to absorb volatility.
Then the filing reveals the other side.
The Commitments Are Also Astonishing
Anthropic needs enormous amounts of compute. The S-1 reports $8.945 billion of compute and infrastructure expense in the first half of 2026 and describes long-term purchase commitments, take-or-pay arrangements and infrastructure relationships extending for years.
Reuters reports compute commitments exceeding $600 billion. The filing itself makes clear that the improved economics still depend on an extraordinary future spending path.
So Anthropic simultaneously has enormous revenue growth, enormous liquidity and enormous future obligations. Those facts belong in the same conversation.
This Is Why “Bubble” Is Such an Unhelpful Binary
People tend to talk about bubbles as though the underlying thing must be fake. It isn't.
The internet bubble did not prove that the internet was useless. Railway booms did not prove that railroads had no economic value. A technology can transform the economy and still attract more capital than the eventual returns justify.
That may be the most important lesson in Anthropic's prospectus. Claude can be extremely useful. Anthropic can become enormously profitable. AI can reshape entire industries. And investors can still overpay for the companies and infrastructure built around it. All four things can be true.
Amazon, Google and the Rest of the Stack Are Everywhere in This Story
The filing also exposes just how intertwined the AI economy has become. Anthropic depends heavily on a limited group of cloud, hardware and distribution partners. Those firms can simultaneously be investors, suppliers, distributors, customers and competitors.
There is nothing inherently improper about that. It may be an extremely efficient business arrangement. But it makes the AI economy difficult to analyze using simple categories like customer, supplier, investor and competitor.
The same companies occupy several categories at once. That matters when trying to understand where losses would land if the cycle turns.
Customer Concentration Is Better Than You Might Expect — But Still Meaningful
Anthropic has an enormous customer base. That reduces some risk. But concentration remains material, and many customers are not locked into contracts for the full life of the infrastructure being built to serve them.
That creates an interesting mismatch. Infrastructure can be committed for many years. Customer demand can change much faster. The business works beautifully if usage keeps expanding. It becomes more difficult if compute commitments are fixed while customers remain flexible.
The Filing Changes My View — Just Not in the Obvious Way
Before the IPO filing, Anthropic was one of the larger unknowns in the AI investment cycle. We knew it was growing. We knew it was raising enormous amounts of money. We knew it had major partnerships. We did not know enough about the operating business to confidently say how much of that infrastructure spending was being supported by actual economic activity.
Now we know more. The revenue is substantially stronger than the private-company opacity suggested. The operating trajectory is much better than the headline net-loss number implies. The balance sheet is enormous.
That makes me less concerned about whether there is real demand for frontier AI. It does not make me less concerned about the scale of the investment cycle surrounding it.
The Bigger Question Is No Longer “Does Anyone Want AI?”
They do.
The more useful question is: How much capital should rationally be invested to meet that demand?
A business can be great. Its stock can still be overpriced. A product can be transformative. Its infrastructure can still be overbuilt. Revenue can grow rapidly. Financing commitments can still assume even faster future growth.
This is why EconRisk separates Demand / ROI from CapEx Sustainability, Financing / Credit and Market Concentration. Anthropic's filing lowers one kind of risk while increasing our confidence in several others.
The IPO Will Become a Price Discovery Event for the Entire AI Economy
Anthropic is applying to trade on Nasdaq under ANT. The final price has not yet been set. That may ultimately be more important than any single number in the prospectus.
Private investors have been valuing leading AI companies using private financing rounds, strategic investments and negotiated transactions. The IPO will let the public market decide what one of those businesses is worth every day.
That gives the AI boom something it has lacked: continuous public price discovery for a frontier-model company.
If investors accept an enormous valuation, that will strengthen confidence in the sector. If they demand a major discount, it will tell us something equally important. Either way, Anthropic becomes a benchmark.
The AI Bubble Argument Just Got Better
Not because Anthropic looks terrible. Quite the opposite.
A credible bubble argument should be able to survive evidence that the underlying technology works.
Anthropic makes that possible. The company is growing spectacularly. The demand is real. The customers are real. The revenue is real. And the economic potential may be enormous.
The remaining question is whether the amount of money being committed around that opportunity is rational relative to the returns it will eventually produce.
That is a much harder question than asking whether AI is useful. It is also the question that matters.
← More commentary