Anthropic's latest quarterly revenue reached 16.8 billion US dollars, with a further 45% quarter-on-quarter increase.

According to Reuters, an investor who attended Anthropic's pre-IPO executive meeting in San Francisco on Wednesday revealed that Anthropic's revenue in the third quarter rose 45% quarter-on-quarter to $16.8 billion. The company also told investors that it achieved profitability in the third quarter on an adjusted basis.

When you break down the numbers, the slope of this growth curve is quite steep, but the slope itself is also changing.

$4.7 billion, $11.6 billion, $16.8 billion: More than tripled in three quarters

The quarterly figures provided by attendees at the Investor Day are as follows: revenue of $4.7 billion in Q1, $11.6 billion in Q2, and $16.8 billion in Q3.

From Q1 to Q2, revenue nearly tripled (+147% quarter-on-quarter); from Q2 to Q3, the growth rate slowed to +45% quarter-on-quarter, but the absolute value still kept rising — Q3 generated $5.2 billion more than Q2, while the increment of Q2 over Q1 was $6.9 billion. The "acceleration" of growth is declining, but the "absolute increment" supported by scale remains at a high level.

Converted based on the single-quarter revenue of $16.8 billion in the third quarter, Anthropic's current annualized revenue run rate has reached around $67 billion. A year ago, the company's total annual revenue was only about $4.6 billion. Reuters reported last month that Anthropic's revenue grew 12 times last year to nearly $4.6 billion, despite an operating loss of more than $8 billion (excluding impairment of various liabilities primarily related to previous financings).

The growth is real, and the cash burn is also real. These two lines are now converging into the same IPO window.

The weight of the phrase "adjusted profitability"

Anthropic told investors that on an adjusted basis, the company achieved profitability in the third quarter.

"Adjusted" is the most critical part of this phrase. It did not state that it was profitable under the GAAP standard, nor did it disclose which specific items were adjusted out. Combined with last year's data, this wording needs to be interpreted in context: in 2025, Anthropic recorded an operating loss of more than $8 billion, and that was the caliber after "excluding impairment of liabilities primarily related to financing".

This means that the company's profitability is improving, but the outside world does not have enough detailed data to verify to what extent the improvement has reached and how far it is from positive free cash flow. The pre-IPO Investor Day is inherently a sales roadshow, and the quantifiable figures usually only select the most convincing part to present.

This is also why the market keeps a close eye on the gap between "adjusted" and "operating basis" — it measures how solid the conversion rate of an AI lab turning revenue into profits is.

The arithmetic problem to be solved for a $2 trillion valuation

Reports say Anthropic is targeting a valuation of more than $2 trillion.

Calculated based on the single-quarter revenue of $16.8 billion in Q3 and the annualized run rate of about $67 billion, the $2 trillion valuation corresponds to a price-to-sales ratio of approximately 30x. If calculated based on the full-year revenue of $4.6 billion last year, the multiple will rise to more than 40x.

This range is not cheap for ordinary software companies, but for AI labs that are "believed to be able to define the next-generation computing platform", it follows a different set of valuation logics. The problem is that the 30x P/S ratio implies a premise: growth can be sustained, and the cost curve can eventually be flattened by revenue.

At the other end of the cost curve is a staggering bill.

A $5.18 trillion ten-year arms race

Anthropic has committed to investing at least $5.18 trillion in cloud computing services, computing power and infrastructure over the next ten years.

Break down this figure: an average of about $518 billion per year, which is already close to 80% of its current annualized revenue run rate (about $670 billion). In other words, the amount of money it promises to spend on "building facilities, purchasing chips, and renting computing power" is already quite close to the amount of money it earns from the market.

This is exactly the root cause of the increasingly strict financial review of AI companies. As AI companies pour capital into chips, data centers and computing power to develop and run increasingly powerful models, the market begins to ask a simple question: When can revenue growth truly cover these expenses?

What Anthropic needs to convince investors is not that "it is growing very fast" — which the figures have already proven — but that "growth can eventually offset the huge cost of building and running AI systems". This is the logical fulcrum for the $2 trillion valuation to hold, and it is also the question that it and peers like OpenAI need to answer together.

Preemptive IPO is an acceptance test

Anthropic's IPO is expected to precede the listing of its rival OpenAI. Reuters reported that Anthropic's public market debut may be postponed to after the US midterm elections in November.

The significance of the preemptive move is not just about "who rings the bell first". It makes Anthropic's listing an early test for investors: Will the public market be willing to pay for AI developers with thin profits but pursuing large-scale applications?

Appearing ahead of OpenAI, Anthropic is effectively testing the waters for the entire "frontier AI lab" track. If the market gives a sufficiently high valuation, the financing window for the players queuing up behind will be opened; if the response is lukewarm, the narrative of the entire track will need to be re-evaluated. This is why signals such as Q3 profitability and controllable costs have been repeatedly emphasized and presented to investors on the eve of the IPO.

Google and Amazon: Backers, suppliers, and also rivals

Anthropic was founded by former OpenAI executives and has become one of the leading generative AI developers with its Claude series of models. Its expansion is supported by Alphabet and Amazon — the two are not only investors, computing power suppliers and important distribution partners, but also competitors in the field of artificial intelligence.

The tension in this relationship is real. Anthropic's computing power largely relies on the clouds of Google and Amazon, and its models are also distributed through the channels of the two companies; but the two are also developing frontier models on their own, making them both its funders and track rivals. Claude is provided to the public through platforms such as Amazon Bedrock and Google Cloud, which on one hand serves as distribution, and on the other hand places its core products on the shelves of potential competitors.

For investors, this means two layers of risks: first, customer and capital concentration — large tech companies are both backers and potential competitors; second, bargaining power — when the computing power supplier and the largest distribution channel are the same entity, Anthropic's say in the cost side will be weakened. A large part of the $5.18 trillion computing power commitment will eventually flow to the cloud businesses of these two companies.

The safety card, and the unsettling results in the test

Anthropic has always highlighted its own characteristics by emphasizing artificial intelligence safety, warning of the dangers posed by increasingly powerful systems, while also competing to develop and commercialize more powerful models.

Its own research has yielded a thought-provoking detail: in the test environment, its AI agents showed disturbing behaviors, including leaking sensitive information and sabotaging code. The company said that no such inconsistent evidence has been found in actual deployments.

The caliber of "existing in tests, not in deployments" is both a reassurance and leaves an open question: to what extent can the gap between the test environment and real deployment represent the safety boundary of model behavior? When Agents start to call tools autonomously and access sensitive data, the red team results inside the lab will be closer to the real location of risks than external declarations. For a company that uses "safety" as its differentiator, how such internal findings are disclosed and governed is itself a page that investors should pay attention to.

Beyond the figures, Anthropic has pushed itself to the doorstep of IPO with a report card of $16.8 billion in a single quarter and a further 45% quarter-on-quarter increase. The business story it wants to tell is clear: extremely fast growth, adjusted profitability, backed by Google and Amazon, and targeting the next-generation computing platform.

But the questions the market needs to answer are more complex. The 30x P/S ratio buys the expectation that "future costs can be flattened by revenue"; the $5.18 trillion ten-year investment buys the bet that "scale can build a moat"; and the gap between "adjusted profitability" and "operating loss of $8 billion" will not show its true shape until the prospectus is disclosed and auditors step in.

Until then, all the figures are still just a presentation at the Investor Day.

This article is from the WeChat Official Account "Tech Business", authored by Tech Business, and authorized for release by 36Kr.