Wall Street doesn't know how to value AI anymore.
Anthropic's growth rate once again exceeded its predictions.
According to investor documents seen by Bloomberg, Anthropic earned more than $11.5 billion in initial revenue in the second quarter of this year, up more than 14 times from $787 million in the same period last year; and more than doubled from $47.3 billion in the first quarter of this year.
Anthropic also recorded a positive adjusted operating profit in the second quarter. The relevant figures are still preliminary and the final results may be adjusted.
Just the day before, Reuters exclusively revealed a more startling figure:Anthropic expects annual revenues of the company to reach $190 billion to $200 billion by 2028.The news came from two people familiar with the financial situation of the company, data that has not been previously reported.
With Anthropic's IPO on the rise, Wall Street has begun to use this "two-year-old Anthropic" to determine how much Anthropic is worth today.
Two years from now, can a company that relies heavily on selling AI modeling capabilities really make $20 billion a year?
The latest disclosure of $11.5 billion in quarterly revenue leaves the seemingly crazy "$20 billion story" not that far away.
Growth is so fast that it can't keep up with traditional valuations
The Anthropic growth curve is already quite rare.
In the second quarter of last year, the company's revenue was just $787 million; it increased to $47.3 million in the first quarter of this year and to more than $11.5 billion in the second quarter.
If we simply annualise the latest quarter's revenue, Anthropic is now equivalent to a company with about $46 billion in annual revenue.
This is in line with the data released by Anthropic in May this year. When the company completed its new round of financing, it said its annual run-rate revenue had surpassed $47 billion.
This scale was built up almost over a year. At the beginning of 2025, Anthropic's annual operating income was only about $1 billion; by August 2025, it had surpassed $5 billion. The company said its annual operating income grew more than 10-fold for three consecutive years.
This year's acceleration is even more evident. Financial projections previously obtained by Reuters show that Anthropic originally expected revenues of at least $10.9 billion in the second quarter of this year and recorded a quarterly operating profit of about $559 million. Initial revenues disclosed today have exceeded $11.5 billion, higher than previously forecasted.
That's why it looks extremely exaggerated that in 2028, "$20 billion in revenue" will begin to enter the Wall Street valuation model.
Based on current annual revenues of approximately $46 billion to $47 billion, Anthropic would need to expand by about four times to reach $200 billion.
It's still a very radical prediction if you look at a mature software company. But on the growth curve of Anthropic over the past few years, things are starting to get complicated.
Wall Street starts pricing for Anthropic two years later
Reuters quoted four people with knowledge as saying that in the pre-introduction valuation discussion of Anthropic, banks and investors are using EV-to-Revenue e, the company's value/revenue multiplier, to directly predict revenues for 2028.
For fast-growing, profitable software companies that have not yet matured, it is not uncommon to value profits by revenue rather than profit.
Why?Anthropic is growing too fast. Investors are also facing a particular problem.
If the profits of today were to put an estimate on Anthropic, a huge amount of GPUs, model training, reasoning, talent and data centers would make it look unusually expensive.
Assuming these infrastructures could ultimately support an AI platform that expands multiple times in scale and keeps increasing profits, today's investment may be just the cost of a period of rapid expansion.
Wall Street is actually betting on a second scenario. According to Reuters, high-growth companies such as Palantir, Cloudflare and SpaceX are currently being cited by investors as being valued by Anthropic.
Palantir: It has become the benchmark for investors when valuing companies with high growth and deep involvement in AI.
Cloudflare: A direct contrast to high-growth software and infrastructure companies.
SpaceX: Provides a case where the valuation of an enterprise can be partially dependent on expectations of its future size rather than on current financial fundamentals.
Some of the fastest-growing companies on the market in recent years have had similar precedents.
Supporters of Cerebras cited the company's earnings forecast for 2028 ahead of its IPO this year. SpaceX even extended its financial forecast to 2029 before launching at a record valuation in June this year.
This also reflects the difficulty of valuing an AI company.
$200 billion, who's going to pay?
Following the release of the Reuters report, the discussion quickly spread to the Reddit technology community.
Supporters argue that from the growth rate of Anthropic over the past three years, the $200 billion is not as incredible as it seems at first glance. The potential market is still huge, especially as enterprise AI is still in its early stages and a large number of companies have not yet really deployed generated AI into their core workflows.
But the questioner asks if the same growth logic is going from billions of dollars to billions of dollars.
Some netizens point out that it's one thing for the market to pay Claude tens of billions of dollars a year; it's another thing for businesses to end up making close to $20 billion.
It has also been suggested that if Anthropic, OpenAI, and other AI giants are expected to generate hundreds of billions of dollars in revenue in the future, which part of the economic system will ultimately shift the huge revenue expected from the entire AI industry?
No matter where it comes from, the $200 billion in revenue will ultimately have to match the actual customer spending of $200 billion.
Anthropic is the biggest risk, AI is getting cheaper
Some users are focusing on open-weight models, local models, and increasingly inexpensive competing products.
Companies may not always need the strongest model in the world, but only one that is low-cost enough to complete the task. Once large amounts of AI capabilities are commercialized, companies will be motivated to look for lower-cost alternatives, and model manufacturers' pricing capabilities will be under pressure.
This is one of the most easily overlooked risks in Anthropic's "$20 billion forecast".
Revenue essentially depends on two variables: use, and price. The market now generally believes that the demand for AI will increase dramatically. But if the price of future model unit capabilities drops as quickly, the explosion in use does not necessarily mean that revenue explodes at the same rate. There may even be a counterintuitive result: the faster AI advances, the more expensive front-end capabilities today, the faster they become the cheap commodities of tomorrow.
What Anthropic really needs to build is not just a model leader, but to embed Claude into the workflow of the company, making it increasingly difficult for customers to replace.
Currently, Claude's growth in the programming and professional workplace is one of Anthropic's most important milestones. Reuters previously noted that corporate customers and the rapid adoption of Claude Code were important drivers of the company's recent revenue surge.
It's not the hardest thing to earn, it's the profits.
In the past, the market has often asked when AI companies can make money. Now the question becomes further: can this profit be sustained and can it grow rapidly as it expands in scale?
One of the biggest differences between AI and traditional SaaS is that every model call requires real computing resources. A traditional software company adds 1 million users and the cost of a marginal server may be low; but large language models generate GPU, power, and data center resources for every token.
Therefore, to support a valuation that is close to a trillion dollars or even higher, Anthropic must also prove that revenues will eventually grow faster than computational costs.
This is also the most central assumption for investors in the Reuters report that the future profits of Anthropic could continue to expand as models, chips and reasoning efficiency improve, and fixed costs such as personnel are weakened by larger revenue scales.
If this assumption is true, then the huge computing power invested today could be similar to Amazon's logistics infrastructure of the early years, for the future to build ahead of schedule. If it is not, the higher the revenue and the higher the computing bills, the harder it will be for AI companies to enjoy double the high valuation of traditional software companies.
At least one thing is already clear: The market has underestimated the growth rate of Anthropic more than once.
It demands that enterprise AI spending continue to grow rapidly; that Claude maintain strong enough product competitiveness; that open models and low-priced competitors cannot quickly squeeze down prices; and that computational costs fall fast enough to ultimately turn revenues into massive profits.
The $1.5 billion in quarterly revenue has proven that Anthropic's ability to sell AI is far stronger than most people thought. But to rise to the trillion-dollar valuation that Wall Street is discussing, it will have to prove a much older business problem: how much money can it ultimately leave behind.