#Gate直通IPO第三期Anthropic What underpins Anthropic’s trillion-dollar valuation?

Anthropic has officially begun its push into the capital markets. Its IPO will provide an important valuation benchmark for the large-model industry. This article examines its profitability, customer mix, and technological moat to explore why it is worth so much.

01 Rapid revenue growth, with profitability beginning to materialize
“Spending heavily on compute without seeing a return” has long been a challenge for all large-model companies. Anthropic is one of the few to break that pattern. Its revenue growth has already reached astonishing levels. IPO filings show that Anthropic generated just $41 million in revenue in 2023; that grew to $386 million in 2024, up 841% year over year, and then to $4.59B in 2025, up 1088%.
Growth accelerated further in 2026. In the second quarter of this year, the company generated $11.56B in revenue, surpassing OpenAI’s approximately $6.7B in revenue for the same period for the first time. Alongside its rapid revenue growth, Anthropic also reached an inflection point in profitability. In the second quarter of this year, Anthropic recorded its first quarterly operating profit, at $974 million.
Although it failed to remain profitable in the third quarter, it at least showed the market that profitability is possible. By comparison, OpenAI, another global AI leader, has continued to grow its revenue but remains deeply unprofitable. The company’s operating losses reached $9.3B and $12.3B in the first and second quarters of 2026, respectively, mainly because it spent heavily on model training and R&D. Even Anthropic, which has already turned a profit, has not fully escaped these pressures. This also shows that profitability for large-model companies always comes up against the hurdle of compute costs. Whether training new models or meeting users’ growing inference needs, these companies must continue to consume huge amounts of compute.
So the key to whether the business can be profitable is whether revenue growth can outpace growth in compute spending. One important reason Anthropic was profitable in the second quarter of 2026 is that its revenue grew faster than its compute costs. In the first half of the year, the company invested $6.2B in training-related compute, but compute spending had already fallen to 55% of revenue, down from as much as 653% two years earlier.
However, this cost structure is not set in stone. If customer demand suddenly surges or a new model needs to be upgraded, compute spending could rise again. Profitability in a single quarter does not mean large-model companies have completely emerged from their cash-burning phase. Whether they can remain profitable still needs to be tested over a longer period.

02 Doubling down on B2B and growing enterprise monetization
Having covered hardware investment, let’s look at the business model. From the outset, Anthropic has focused firmly on enterprise services and the developer ecosystem, giving it far stronger user retention and product repurchase rates than AI products focused mainly on consumers. Its revenue mix has taken shape as a model built primarily on model usage, with subscriptions as a secondary source.
According to IPO filings, about 75% to 85% of Anthropic’s revenue in the first three quarters of 2026 came from enterprise APIs and usage-based billing, while consumer subscriptions accounted for only about 15%.
How much revenue do enterprise customers actually contribute?
As of the end of September 2026, Anthropic had 11.9k organizational customers that had contributed more than $100k in revenue over the previous 12 months, up from just 950 a year earlier. Of these, more than 1,900 customers contributed over $1M annually, and around 250 contributed over $10M. These high-spending customers have become the mainstay of Anthropic’s revenue. Over the previous 12 months, customers contributing more than $100k annually generated a combined $26.1B, accounting for 74% of total revenue during the same period. Growing the customer base is only the first step; whether existing customers continue to increase their purchases is a more important indicator of revenue quality.
Anthropic most recently disclosed a net revenue retention rate of about 900%. By its calculation, excluding new customers, the total amount existing customers paid over the course of a year reached nine times the amount they paid in the same period a year earlier. This reflects companies embedding the models more deeply into their workflows and steadily increasing usage.
In comparison, consumer willingness to pay is prone to fluctuation, and the marginal cost of serving consumers is high. With model inference costs soaring, massive traffic could become a burden on profits. That is why OpenAI, which initially made a major push into the mass consumer market, has also begun accelerating its drive to win enterprise customers.
OpenAI’s chief revenue officer revealed in April this year that enterprise revenue already accounted for more than 40% of the company’s total revenue, and forecast that it would reach a similar scale to its consumer business by year-end. By comparison, Anthropic’s share had already reached around 80% over the same period. This is why the capital markets are paying attention to B2B businesses. For SaaS companies, existing customers’ willingness to keep paying and increase their purchases shows that the product is creating real value and that future cash flows will be more stable—naturally leading the capital markets to assign a higher valuation.

03 Only technological leadership can sustain a high valuation
However, the revenue growth and B2B customer advantages mentioned above do not mean Anthropic can maintain its lead indefinitely.
The AI industry is changing rapidly, and technology is advancing by leaps and bounds. A single model iteration could shift the competitive landscape or even overturn the existing valuation logic. Ultimately, technological capability is what determines whether customers stay or leave.
IPO filings show that new model versions released in the past six months contributed more than 80% of revenue in the third quarter of 2026. In other words, customers follow improvements in model capabilities and actively switch to new versions, with those improvements directly driving growth in paid usage. Even though Anthropic raised its API prices by 2.3–4.5x in 2026, there was no significant drop in enterprise customer retention. Customers’ acceptance of the price increases suggests that Claude is difficult to replace for coding agents and complex, long-horizon tasks.
Of course, technological leadership comes at a price. Continuously iterating on models requires ongoing investment in compute and talent, and those costs could put pressure on future profitability. But that does not stop Anthropic from being a rare benchmark for the AI industry.
Today, it is hard to find a company that can be directly compared with it. It has both leading model capabilities and a proven enterprise payment and commercialization model. Traditional valuation methods for internet and SaaS companies are difficult to apply in full. What’s more, competition across the AI market is far from saturated. Anthropic still has many potential customers, leaving considerable room for future revenue growth.
The capital markets are willing to pay a premium for this scarcity and growth potential.

So, do you think Anthropic is worth a trillion dollars? Let’s talk~