AI trading, the growth rate has started to fall short of expectations.

On October 14, per FT, Anthropic disclosed partial financial figures to IPO investors, showing that the company's revenue in the third quarter reached 168 billion US dollars, a quarter-on-quarter increase of roughly 46%, marking the second consecutive quarter of adjusted profitability.

While the growth rate appears relatively high, Anthropic's revenue growth has actually slowed down significantly. In the first quarter of this year, the company's revenue stood at around 48 billion US dollars, and rose to 115 billion US dollars in the second quarter, representing a quarter-on-quarter growth of more than 100%. The quarter-on-quarter growth rate in the third quarter fell below 50%.

Research institution SemiAnalysis estimates based on the latest quarterly data that Anthropic's current annualized revenue is approximately 740 billion US dollars. Previously, some investors had expected the company's annualized revenue to far exceed 1 trillion US dollars by the end of the year.

However, Reuters cited investors attending the meeting as saying that Anthropic still expects its annualized revenue to reach 1.1 trillion US dollars by the end of the year.

Six days ago, OpenAI also disclosed a revenue figure that was "lower than previous market expectations".

On October 8, FT cited financial documents obtained by investors and reported that as of the end of September, OpenAI's annualized revenue was close to 500 billion US dollars, roughly 200 billion US dollars less than the previously widely circulated market figure of 700 billion US dollars.

On the day the news was released, NVIDIA fell 2.9%, AMD fell 3.9%, and Broadcom fell 4.3%. As of October 14, adjustments also took place in the Chinese market, with the STAR 50 Index down 3.29% on the 14th.

The scenario that the capital market was most worried about has arrived: the revenue of leading AI manufacturers falls short of expectations. The valuations of industrial chain enterprises that previously benefited from capital expenditure have started to correct. Moreover, from model companies to upstream chip suppliers, the market has begun to re-evaluate the sustainability of revenue growth and capital expenditure in the AI industrial chain, as well as valuation standards.

As of press time on the 15th, the STAR 50 Index fell 2.9%.

Anthropic's latest financial figures: growth rate lower than expectations

In the first half of this year, the market expected that Anthropic's IPO valuation might set a new listing record, with a previous valuation of 965 billion US dollars. Related information showed that the company planned to go public in November and seek a higher valuation.

Prior to its listing, Anthropic demonstrated its achieved commercialization scale to investors. In addition to the 168 billion US dollars revenue in the third quarter, the company also disclosed profitability indicators.

Per FT, Anthropic has achieved positive cash flow, with an adjusted operating profit margin of 11% in the third quarter. "For every 1 US dollar invested in computing resource costs, the company can generate 1.70 US dollars in revenue". This means that Anthropic's model service revenue can already cover the corresponding computing resource costs, and achieve operating profitability under the adjusted caliber. However, these figures do not yet reflect the full capital expenditure and all accounting costs.

For Anthropic, which has a valuation close to one trillion US dollars, the market's focus has shifted to future growth.

In the first half of this year, Claude's competitiveness in fields such as programming helped Anthropic rapidly expand its enterprise customer base. But entering the second half of the year, the company's revenue growth began to be affected by competition.

In July this year, OpenAI launched a new model, which regained its competitiveness in some application scenarios. At the same time, Anthropic's key customers Meta and Microsoft began to use more self-developed AI tools, and domestic open-source model manufacturers are also competing for the market through low prices.

Joey Brookhart, an analyst at SemiAnalysis, said that over the past three months, investors have been lowering their annualized revenue expectations for Anthropic. In the first half of this year, Anthropic maintained a leading position in some markets, but now OpenAI has re-emerged as a major competitor.

In addition, Anthropic has also actively reduced part of its revenue sources. Anthropic told investors that revenue related to model distillation accounts for 5% to 7% of total revenue, and the company has significantly restricted the relevant usage in the third quarter. The model distillation mentioned here refers to other model developers using the output results generated by advanced models to train their own models. Due to the need for a large number of API calls, such behavior previously contributed revenue to Anthropic.

But the company's long-term capital expenditure plan is still huge. Per Reuters, Anthropic's committed expenditure on cloud services, computing resources and infrastructure over the next ten years exceeds 5.18 trillion US dollars.

It is worth noting that Anthropic's management also reminded investors that the current profitability status may change as the company continues to invest in computing power and biomedical research. Calculated based on the valuation of 965 billion US dollars and the 740 billion US dollars annualized revenue estimated by SemiAnalysis, Anthropic's valuation is approximately 13 times its annualized revenue.

OpenAI's revenue also "fell short of expectations" before

Compared with Anthropic, the controversy triggered by OpenAI's latest financial data focuses on the revenue accounting method.

On October 8, per FT, OpenAI disclosed to investors that as of the end of September, the company's annualized revenue was close to 500 billion US dollars. However, the market had previously widely estimated that OpenAI's annualized revenue was close to 700 billion US dollars. In other words, this was also "lower than expectations".

It should be pointed out that the 200 billion US dollars gap between the two figures comes from the different revenue recognition methods adopted by OpenAI and Anthropic. Anthropic includes sales revenue realized through cloud partners such as AWS and Google Cloud in its own revenue. OpenAI does not adopt the exact same processing method. Previously, in order to compare the two companies, investors adjusted OpenAI's revenue figures, forming the 700 billion US dollars market expectation.

From the above caliber, perhaps the change in revenue data does not mean that OpenAI suddenly lost 200 billion US dollars in revenue. The revenue disclosed by the company to investors is still growing, with an increase of more than 70%.

But the capital market's reaction to this news was relatively direct.

On October 8, major chip stocks such as NVIDIA (NVDA.US), AMD (AMD.US) and Broadcom (AVGO.US) fell collectively. The Nasdaq Index fell by about 1.3% that day, and the Philadelphia Semiconductor Index fell by more than 3%.

This reaction is related to OpenAI's position in the AI industrial chain. Over the past few years, OpenAI has been an important driver of the growth in global AI computing power demand -- the company's expanding financing scale, computing power procurement agreements and data center construction plans have provided order expectations for upstream chip suppliers. At the same time, the financial data of model companies themselves have long lacked the full disclosure required by the public market. Investors mostly estimate the growth space of the company based on annualized revenue, user volume and capital expenditure plans.

When there is a 200 billion US dollars difference in the revenue statistical caliber, the market has begun to re-examine the basis of these estimates. This also makes OpenAI's future capital expenditure a key focus for investors.

Moreover, model training and reasoning require continuous procurement of computing resources. OpenAI also needs to face the decline in model prices, as well as competition from Anthropic, domestic open-source models and self-developed models of large technology companies.

Synchronous adjustment of AI assets from US stock market to STAR Market

The market's concerns about AI investment have extended from revenue expectations to financing costs.

On October 13, the Philadelphia Semiconductor Index fell for the fifth consecutive trading day. On the 15th, Micron Technology (MU.US) fell 4.5%, storage-related company SanDisk (SNDK.US) fell 4.3%, Seagate Technology (STX.US) fell 5.9%, and Western Digital (WDC.US) fell 5.8%.

In addition to changes in AI capital expenditure expectations, the technological progress in storage made by Chinese storage chip manufacturer Changxin Technology (688825.SH) has also triggered market concerns about intensified competition in the global storage chip sector.

On October 14, the US announced that the CPI in September rose 3.7% year-on-year, higher than market expectations. The inflation data increased the market's concerns about the Federal Reserve raising interest rates further. On the same day, the Nasdaq Index fell 1.88%, and the S&P 500 Index fell 1.11%.

However, AI infrastructure companies saw even larger declines. CoreWeave (CRWV.US) fell by about 16%, Nebius (NBIS.US) fell by about 11%, Oracle (ORCL.US) fell by about 9%, and Dell Technologies (DELL.US) fell by about 11%. According to Reuters statistics, the four companies lost a total of about 800 billion US dollars in market value. The business of these enterprises relies on large-scale computing power infrastructure construction. Rising interest rates will directly increase financing costs and also suppress the return on investment of some projects. CoreWeave's bond prices have already reflected this pressure. On October 14, the yield of some of its bonds rose to around 15%, as investors demanded higher risk compensation.

The Chinese market has also undergone adjustments.

On October 14, the STAR 50 Index fell 3.29%; the STAR Composite Index fell 2.97%. Among the 617 normally traded stocks on the STAR Market, only 58 rose. Domestic AI chip companies still benefit from domestic substitution, and the driving factors for their business growth are different from those of US chip enterprises. But over the past period of time, the valuations of these companies have also been boosted by expectations of growing demand for AI computing power.

With the adjustment of global AI assets, the A-share technology sector has also begun to face pressure on valuation and capital supply.

The AI narrative is starting to change: entering the financial verification phase

To sum up, over the past few years, AI trading has been built on continuously expanding capital expenditure.

Model companies such as OpenAI and Anthropic obtain funds through financing, and invest the funds in GPUs, data centers and cloud computing services. NVIDIA, server, storage and data center enterprises obtain orders through these investments.

The capital expenditure plans of model companies have thus become an important basis for the market to predict the revenue of upstream enterprises.

But this growth relationship has a premise: model companies ultimately need to bear the increasingly high computing costs through commercial revenue.

The latest financial data disclosed by Anthropic and OpenAI has started to make this issue more specific. Anthropic has achieved relatively high revenue and adjusted profitability, but its growth rate is lower than the previous level. OpenAI is still growing rapidly, but its latest revenue statistics have a gap with previous market expectations. Both companies still need to continuously invest in computing resources to maintain model competitiveness.

At the same time, competition in model prices is intensifying. The progress of customer self-developed models and open-source models has also begun to affect the revenue growth of leading model companies.

These changes have made it more difficult for the market to judge the long-term profitability of model companies. For upstream computing power enterprises, order growth still has support, but valuation needs to take into account the capital return of customers. As interest rates rise, enterprises that rely on external financing to build data centers also need to bear higher capital costs.

Judging from the currently disclosed data, AI commercialization is still expanding, and the revenue growth of leading model companies has not stopped. Some chip enterprises still maintain a high profit growth rate.

But the capital market has begun to raise requirements for AI assets.

In the past, when model companies announced larger financing and computing power procurement plans, it might push up the valuations of upstream enterprises. Now, investors are starting to pay attention to the actual revenue, profits and cash flow of model companies, and judge the sustainability of capital expenditure based on these indicators.

Soon after, Anthropic's IPO launch will further test the market's valuation expectations for leading model companies, which will also simultaneously affect the valuations of domestic large model listed companies and to-be-listed companies.

AI trading has entered the financial verification phase. Model capabilities, capital expenditure and computing power orders are still important, and revenue growth and capital return will be given higher weight. The financial performance of Anthropic and OpenAI shows that the AI industry has generated considerable commercial revenue. But this revenue still needs to support trillions of US dollars of computing resource investment in the future.

The years-long AI capital expenditure narrative is facing new constraints.

This article is from the WeChat official account "Bright Company" (ID: suchbright), author: Editor on duty 24/7, published with authorization from 36Kr.