Anthropic's $2 trillion IPO questioned: computing power promises and risk of revenue mismatch

According to Woofun AI, when Anthropic is about to begin the IPO process, a wave of market questions suddenly soared. Independent research firm New Constructs released a report entitled “The Most Ridiculous IPO 2026,” pointing out that its valuation logic is fundamentally flawed. The core conflict focuses on the serious mismatch between the elasticity of customer revenue and the rigidity of long-term computing power promises. This structural risk is causing severe shocks in the global technology stock market. As the AI industry chain sell-off spreads, Anthropic's fund-raising results are not only about its own R&D and procurement capabilities, but will also reshape semiconductor investors' expectations about the stability of downstream orders.

Behind the differences in valuation is a fierce game between financial models and growth expectations. The valuation given by New Constructs was only $150 billion, about 93% lower than the market's expected target of $2 trillion. This huge margin stemmed from deep doubts about revenue sustainability. Anthropic's annualized recurring revenue (ARR) is expected to reach $85 billion in September. Based on this calculation, the $2 trillion valuation corresponds to nearly 20 times the market sales ratio (P/S), which is extremely expensive.

However, Rob Li (Rob Li), managing director of Amont Partners, pointed out that the rationality of the valuation depends on the growth rate: if ARR reaches $100 billion at the end of the year and doubles to $200 billion next year, the long-term P/S will drop 7-8 times, entering a reasonable range.

Notably, Anthropic's ARR soared from $9 billion in January to $65 billion in July, growing more than sevenfold during the year. According to data compiled by Woofun AI, if New Constructs' $150 billion valuation is divided by $65-80 billion ARR, P/S is about 2-2.3 times, which is 4.2 times lower than the overall median SaaS and 4.7 times the median for generative AI theme companies. For Anthropic, which has high growth and is unprofitable, the price-earnings ratio (P/E) fails due to strategically depressed or sharp fluctuations, and P/S is a more appropriate measure, but the key is whether the growth rate can continue to absorb high valuations.

Historical cases provide a multi-dimensional reference to current valuation disputes. In March 2000, Cisco's (CSCO.US) P/S was 30 times as high. Despite revenue growth, the stock price plummeted 86%, not returning to the starting point for many years, warning of the risk of a mismatch between high prices and growth. Conversely, at the end of 1999, Amazon (AMZN.US) P/S lost about 20 times more and continued to lose money. Although the stock price fell 95%, it became a huge winner for a long time, proving the rationality of high valuations under long-term growth. The 2004 IPO of Google (GOOGL.US) P/S was about 7 times higher. Although it was high at the time, the subsequent high increase in revenue and profit made the valuation low. These cases show that success depends on the company's ability to verify initial valuations through continued high growth, rather than simply looking at multiples. Although Anthropic has a strong position in the model field, the challenge it faces is the mismatch between the two ends of income and expenditure. This makes its valuation logic more complicated than traditional SaaS companies, and needs to be analyzed in conjunction with specific business models.

The pain point of the business model is a structural mismatch between rigid spending on computing power and flexible customer demand. Major customers, such as Microsoft (MSFT.US) and Meta (META.US), which consume more than 1 billion US dollars per year, have publicly stated that they will reduce the use of external models and increase the use of their own models to cut expenses. The two giants contribute more than a quarter of Anthropic's revenue and form the core revenue base, but the cooperation model has always been short-term on-demand procurement, with no lock-up, no price lock-up, and no long-term performance restrictions.

This means that enterprise customers can switch models, reduce calls, or cut budgets at any time, while Anthropic's agreed computing power payments are difficult to reduce in the same proportion. If the additional capacity is not fully utilized, the company still has to pay fees but is unable to obtain expected revenue, leading to a reduction in cash reserves. The mismatch between time spent on additional computing power and contract expiration time further determined that Anthropic must prepare sufficient funds at a specific point in time. This mismatch in the pace of income and expenditure poses a core risk at the operating level.

Macro headwinds and disturbances in market sentiment have exacerbated short-term fluctuations. The war in the Middle East is unresolved. This week's US CPI data greatly exceeded expectations. Expectations of interest rate hikes soared, and interest rate sensitive technology stocks were clearly under pressure. Rob Li pointed out that these challenges are often the reason for the market to short, and it is difficult to falsify in the short term. Hedge fund transactions based on momentum amplify fluctuations in the AI industry chain, and monthly revenue data that falls short of expectations can easily cause panic. Nie Yixiang, head of investment at Fidelity Funds, said that in a high interest rate environment, growth stocks are suppressed as 'long-term bonds', and bear markets pay more attention to cash flow and balance sheets. Early investor selling pressure cannot be ignored either: in 2023, the company was valued at only a few billion dollars, but today it has reached 2 trillion dollars. These investors have a strong incentive to sell stocks after the ban period is over. This is a structural fact, not a question of whether the company is excellent. This can be seen in the fluctuations of SpaceX (SPCX.US).

Future growth verification points focus on enterprise implementation and new industry expansion. Rob Li believes that the key points of observation are additional purchases from existing customers, acquisition of new customers, and expansion of new industries such as healthcare and insurance. Enterprise implementation is the core of sustainability, and whether end customers such as Starbucks (SBUX.US) and Walmart (WMT.US) are willing to continue to pay and receive positive returns is critical. Currently, most companies are in the early stages of testing, and financial institutions have yet to adopt it on a large scale due to security issues. Investors need to observe actual consumption and payment intentions through intermediary service providers. Taking Salesforce (CRM.US) as an example, the growth rate of enterprise spending increased from 20% at the beginning of the year to 40% expected at the end of the year, showing significant growth.

Furthermore, cloud vendors have a “symbiotic relationship” with AI models. Microsoft (MSFT.US), Amazon (AMZN.US), Google (GOOGL.US), etc. profit by providing computing power infrastructure, and the boom in AI applications directly drives the growth of cloud service revenue. If AI requirements are falsified, cloud businesses will be hit hard, and vice versa.

Although cloud vendors may develop their own models to reduce costs, the overall benefits are tied to the expansion of the AI ecosystem, and their ability to apply their own models is still questionable, providing a buffer for model companies such as Anthropic.

In the hardware segment of the industrial chain, storage and optical modules face different risk comparisons. Currently, GPUs, semiconductors, optical modules, etc. are all 'bursting' or in short supply, but investors prefer to 'look forward'. Storage was the most stressful part, but the stock prices of Hynix (SKHY.US) and Samsung fell close to 30%. Rob Li pointed out that the margin of storage security is the highest, but market hesitation stems from its strong cyclical nature and the historical shadow that customers can reduce procurement volume at any time.

However, AI has partially ironed out the memory cycle, and now the memory industry has established more mature guarantee mechanisms, such as requiring customers to pay a 20%-30% advance payment or deposit. Even if the contract is defaulted, suppliers can still keep part of the funds to reduce tail risk. In contrast, although there is a shortage of sectors such as optical modules in the US, the valuation is extremely high and lacks cash reserve protection of the same scale. If the bubble bursts, the downside risk is even greater. Morgan Stanley (MS.US) mentioned that storage companies' inflection points are difficult to predict. The previous cycle ended in 2000, 2016, 2019, and 2022, all due to supply catching up and cloud vendors suspending procurement. Considering the time it takes to expand production, a similar situation is unlikely to occur for most of 2026 and 2027. Other AI hardware aspects lack a prepayment mechanism, and investors are overshadowed by past “supply cuts” and are skeptical about sustainability, which may explain the recent sharp decline in the AI industry chain.

The market cycle summary and investment outlook show that despite strong fundamentals of the AI industry chain, investors will stop and wait while the macro headwinds are not over. In every round of the cycle, the theory supports the driving force during the sharp rise; during the sharp decline, any kind of trend becomes the trigger, and the market spontaneously explores the reason. Anthropic's IPO controversy is not only a valuation issue, but also a stress test for the sustainability of its business model. As the situation gradually becomes clear, the progress of enterprise implementation, changes in the macro environment, and the pace of early investor sell-off will become key variables.