The market appears to be losing its appetite for inflated valuations.

Author: He Jian

Editor: Zhang Wen

Cover Story: Anthropic

Barring any unforeseen circumstances, we could witness the biggest initial public offering (IPO) ever by the year's end.

Multiple media reports indicate that Anthropic is aiming for a staggering $2 trillion valuation as it prepares to go public, with plans to raise around $100 billion. To put this into perspective, in June this year, Elon Musk's combined valuation for SpaceX and xAI post-merger was a mere $1.77 trillion.

However, as the listing date nears, doubts have started to creep in regarding Anthropic's lofty valuation. During a recent investor day, some investors began assessing Anthropic at a $1.5 trillion valuation.

The primary catalyst for this skepticism is Anthropic's decelerating growth. Although Anthropic announced during the investor day that its revenue growth rate in the third quarter remained a robust 45% quarter-over-quarter, with quarterly revenue more than tripling compared to the entire previous year, this still marked a significant slowdown from the 147% growth rate in the second quarter, leaving some investors underwhelmed.

Wall Street has pinned high hopes on Anthropic's IPO, viewing it as a bellwether for the entire AI industry. A lackluster debut could dampen the listing prospects of other firms in the sector. Notably, OpenAI has already postponed its IPO until next year.

To bolster confidence ahead of its IPO, Anthropic has been sharing positive news. During the investor day, the company revealed that it achieved adjusted profitability in the third quarter. Previous media reports had indicated that Anthropic's operating loss exceeded $8 billion last year.

To date, Anthropic has completed nearly 20 rounds of financing, amassing a total of up to $125 billion. In May this year, it announced the completion of a Series H funding round worth $65 billion, with a post-investment valuation of $965 billion.

Growth Slows Down

For a startup established just five years ago, Anthropic's growth trajectory has been nothing short of miraculous.

Last year, Anthropic's annual revenue stood at a mere $4.6 billion. However, in the first quarter of this year alone, its quarterly revenue surged to $4.7 billion, surpassing the total revenue of the entire previous year. In the second quarter, revenue further climbed to $11.6 billion, marking a 147% quarter-over-quarter increase. Within six months, the company's revenue had more than tripled compared to the previous year.

The explosive popularity of Claude Code played a pivotal role in this growth. Anthropic officially launched Claude Code to the public in May last year, targeting programming scenarios, and it quickly gained traction from late last year onwards.

In February this year, Anthropic disclosed that Claude Code's annualized revenue had surpassed $2.5 billion, more than doubling since the start of the year. Much of this growth came from B-end clients, with the number of enterprise subscriptions increasing fourfold, and enterprise usage contributing over half of the revenue.

Claude Code

The number of large clients is also on the rise. In February this year, Anthropic boasted over 500 enterprise clients with annualized spending exceeding $1 million, a number that surpassed 1,000 by early April. Anthropic's annualized revenue also jumped from approximately $9 billion at the end of last year to over $30 billion by early April, further surpassing $47 billion by early May.

During this period, Silicon Valley tech companies were embroiled in a Token maxxing craze, with firms like Microsoft, Meta, and Amazon implementing internal Token rankings to encourage employees to use AI more for work. However, these measures also led to significant unnecessary computing power consumption, with some employees' personal Token consumption reaching over 280 billion in a single month.

Anthropic benefited greatly from this trend and even faced a computing power shortage at one point. To meet demand, Anthropic signed large computing power agreements with Amazon, Google, Broadcom, Microsoft, and NVIDIA, and collaborated with Fluidstack to build data centers. In May this year, Anthropic also leased all the computing power of SpaceX's Colossus 1 data center.

However, entering the third quarter, Anthropic's growth significantly slowed down, with the revenue growth rate dropping to approximately 45% quarter-over-quarter, and the quarterly revenue increase falling from approximately $6.9 billion in the second quarter to about $5.2 billion.

Anthropic's annualized revenue growth also appears to fall short of market expectations. Previously, some investors had optimistically estimated that Anthropic's annualized revenue by the end of the year could far exceed $100 billion. However, according to estimates by research firm SemiAnalysis, Anthropic's current annualized revenue is approximately $74 billion.

Anthropic stated that part of the revenue decline was due to strengthened controls over distillation. Previously, Anthropic had accused other AI companies of distilling its models, but these distillations also brought in significant revenue. According to the Financial Times, Anthropic revealed that distillation-related revenue accounted for approximately 5%-7% of its total revenue.

Simply attributing the decline to distillation does not seem sufficient to explain Anthropic's sharply falling quarterly growth rate. Some market analysts believe that competitive pressure from OpenAI and a plethora of Chinese open-source models may be the direct reasons for Anthropic's slowing growth.

Since the start of the year, OpenAI has continuously updated its models and Codex programming tools, diverting growth from Anthropic's fastest-growing programming market. At the same time, cheaper Chinese open-source models have further increased price pressure, forcing Anthropic and OpenAI to lower their prices in response.

SemiAnalysis analyst Joey Brookhart told the Financial Times that over the past three months, investors have been lowering their expectations for Anthropic's annualized revenue. "Anthropic is now in a duopoly situation, rather than the dominant position it enjoyed in the first half of the year," he said.

Even Larger Computing Power Expenditures

Unlike in the past, when Anthropic's management always emphasized the dangers and concerns of AI on various occasions, at this investor day event, Anthropic's CEO Dario Amodei and CFO Krishna Rao focused on introducing the company's operations and profitability.

Anthropic CEO Dario Amodei

In addition to revealing that the company achieved adjusted profitability in the third quarter, Anthropic also achieved positive cash flow. For every dollar invested in computing power, they generated $1.7 in revenue. After excluding equity incentive expenses, the company's adjusted operating profit margin for the quarter was 11%.

A year ago, Anthropic's revenue was far from sufficient to cover its expenses. According to the prospectus obtained by Reuters, Anthropic's revenue last year was approximately $4.6 billion, with operating expenses of approximately $12.65 billion, resulting in an operating loss of over $8 billion. Among these, computing power and infrastructure expenditures were approximately $7.3 billion, higher than the total revenue for the year.

Computing power is an unavoidable cost black hole for all AI companies. To maintain technological leadership, large model companies have to invest heavily in computing power training costs, and as user scale and usage volume grow, inference expenditures also continue to increase. Among the domestically listed Zhipu and Minimax, over 70% of R&D investment is used for computing power.

Anthropic's computing power expenditures are even higher. According to Reuters, Anthropic disclosed in its prospectus that its cloud services, computing power, and infrastructure expenditure commitments with six partners over the next decade will reach at least $518 billion. Approximately 80% of these commitments are non-cancelable or require payment regardless of usage.

Google and Amazon account for the majority of these commitments. Anthropic has committed to paying Google at least $111.1 billion from April 2026 to July 2033 and Amazon $110 billion from May 2026 to April 2036. According to the contracts, even if actual usage expenditures are insufficient, the company needs to make up the difference.

The agreement with Microsoft involves $31.4 billion, with payments continuing from November this year to May 2033. In addition, Anthropic also has approximately $161.2 billion in equipment leasing obligations related to Broadcom, most of which are also non-cancelable.

Anthropic attempted to explain to investors the necessity of its massive computing power expenditures. Anthropic stated that access to computing power is becoming a key constraining factor for the development of artificial intelligence, and future demand for advanced AI systems is likely to exceed available supply and will be "primarily limited by the availability of computing power."

Even with the cumulative funding of approximately $125 billion obtained through financing so far, it is still insufficient to cover the almost astronomical computing power bills. According to Anthropic's prospectus, as of December 31 last year, Anthropic's total cash, cash equivalents, and short-term investments were only $20.28 billion.

After the record-breaking $65 billion Series H funding round, for Anthropic to obtain more financial support, an IPO is almost the only option. According to previous media reports, Anthropic plans to raise $100 billion, which would make it the largest global IPO in history.

However, for Anthropic, whose growth is currently slowing down, the market no longer seems willing to offer higher premiums.

All rights reserved by Shanshang. No reproduction without authorization.