CoinW Research Institute
Summary
Anthropic is sprinting towards an IPO, with a potential valuation exceeding $2 trillion. According to Reuters citing its confidential prospectus, the company's revenue in 2025 is approximately $4.59 billion, and operational losses are about $8.06 billion; however, after entering 2026, the commercialization of Claude significantly accelerates. The company’s annualized revenue increased from about $14 billion in February and over $47 billion in May to over $65 billion by the end of July; by September, media reports revealed that its annualized revenue had surpassed $100 billion. Claude Code, enterprise APIs, and cloud platform distribution are driving Anthropic’s transition from an AI assistant to enterprise production tools, prompting the market to begin trading its future billion-dollar-level revenue ahead of time. However, behind this high growth lies another set of figures. Anthropic's future cloud computing, computing power, and infrastructure-related contract obligations total about $518 billion, of which approximately 80% involves non-cancelable or minimum procurement and payment arrangements. Whether Anthropic can sustain a higher valuation ultimately depends on whether revenue growth can continue to outpace computing power costs and truly convert into profits and cash flow.
1. Claude opens the books for the first time: Why is the market starting to discuss $2 trillion?
Anthropic is shifting from being one of the most sought-after AI companies in the primary market to the public market. In June of this year, the company secretly submitted an IPO application draft to the U.S. Securities and Exchange Commission (SEC); on September 28, Reuters first disclosed core data from its confidential prospectus, laying out Claude's revenue, losses, and computing costs before the market more comprehensively for the first time. Currently, the issuance scale, pricing, and official listing time are still undetermined. The most eye-catching figures are a set of sharply contrasting numbers. In 2025, Anthropic’s revenue is approximately $4.59 billion, nearly 12 times that of the previous year; but during the same period, operational losses reach about $8.06 billion, of which computing power and infrastructure expenses alone amount to about $7.33 billion. The GAAP net loss disclosed by the company is closer to $42 billion; however, about $34 billion mainly comes from the changes in the liability value of financing tools before converting to equity and other non-cash accounting expenses, which does not mean that Anthropic actually burned $42 billion in a year.
Compared to the more striking net loss figure, the operational loss of over $8 billion better reflects the actual operational pressure faced by the company at that time. Yet, if we only look at the 2025 books, it is hard to explain why the market has already begun discussions about $2 trillion. The real turning point occurred in May this year. Anthropic completed a $65 billion Series H financing, with a post-money valuation reaching $965 billion, more than doubling from $380 billion in February. Just a few months later, the company began preparing for an IPO that may reach $100 billion in scale, further raising its potential valuation to about $2 trillion, with Nvidia reportedly considering becoming an anchor investor for up to $10 billion. Meanwhile, the U.S. IPO market has clearly rebounded this year, with financing reaching a record $137 billion by the end of August. This means that since May, Anthropic's rapid valuation increase has resulted not just from Claude's revenue growth but also from the scarcity of leading AI assets, continuous influx of capital from large technology and infrastructure companies, and the reopening of financing windows in the public market.
This also changes the way we understand Anthropic's valuation. Although the public market is receiving historical financial data from 2025, what is really being traded is the anticipated growth over the next several years. Reuters reports that Anthropic expects revenue to reach approximately $190 billion to $200 billion in 2028, and the close to $2 trillion valuation has effectively anticipated a significant portion of this future growth. If Anthropic aims to maintain a high valuation after going public, the market will ultimately look not just at the revenue for 2028 but at whether Claude can continue to maintain a scale close to $100 billion around 2030, while gradually releasing profits and cash flow. Therefore, the real change that has taken place at Anthropic since May is not just "revenue is growing faster," but that Claude's commercialization, strategic capital, and public market expectations have all been repriced simultaneously. What needs to be answered next is not just how much Claude has made, but what kind of business it has found that makes capital willing to pay so quickly for a more distant future.
Figure 1: Anthropic's financial statements for 2025. Source: CoinW Research Institute
2. Claude's annualized revenue exceeds $65 billion, where is the money coming from?
Entering 2026, Anthropic's revenue growth rate has clearly accelerated. In February, the company’s annualized revenue was approximately $14 billion; by May, it had surpassed $47 billion; and by the end of July, it further broke through $65 billion. By September, media reports disclosed that its annualized revenue had exceeded $100 billion, growing about 50% from two months ago. The term "annualized revenue" (Run-rate Revenue) here does not refer to the total revenue already in hand for the year, but is an annual scale projected based on recent revenue speed. For a high-growth company like Anthropic, it better reflects the current business status than the historical financial reports from 2025, but it is also subject to short-term growth, revenue recognition methods, and changes in customer demand, and cannot be directly equated to future revenue that will definitely be achieved.
The key to this round of growth is that Claude is transitioning from an AI assistant to a production tool that enterprises are truly willing to pay for continuously. The most typical example is Claude Code. It does not just answer "how to write this code," but can also directly participate in the development process, read and modify code, troubleshoot issues, and execute tasks. In February of this year, the annualized revenue from Claude Code had already surpassed $2.5 billion, with enterprise clients contributing more than half of that. The depth of usage by enterprise clients is also increasing. The number of Claude clients spending over $100,000 a year has increased sevenfold in one year, and the number of clients with annualized spending exceeding $1 million now exceeds 500. Among the top ten companies in the Fortune 500, eight are already Claude clients. This indicates that Anthropic's revenue growth comes not only from more users trying Claude, but also from large enterprises allocating more work and budgets to AI. More importantly, these clients are not just purchasing a single product. Enterprises may initially integrate Claude into their applications via API and may start using Claude Code or Claude for Work, gradually expanding to more departments and workflows. Anthropic has also entered IT environments already in use by enterprises through cloud platforms like AWS, Google Cloud, and Microsoft Azure, with over 100,000 clients currently running Claude via Amazon Bedrock.
However, merely looking at the rapid growth of annualized revenue is insufficient to understand the entire growth story of Anthropic. The continuous influx of capital mentioned in the previous chapter not only boosts valuations but also participates directly in business expansion. Anthropic needs massive funds to purchase GPUs and cloud computing resources; companies like AWS and Google are both investors and infrastructure providers, as well as important channels for Claude to enter the enterprise market. Capital, computing power, channels, and revenue thus form a mutually reinforcing cycle: more funds support more computing power and model iterations; stronger products drive enterprise adoption and revenue growth; and higher revenue in turn provides support for the next round of financing and valuation. From this perspective, Anthropic is not merely a valuation story driven by revenue growth. It has a capital cycle similar to the common “capital influx - resource concentration - valuation increase - entry into the secondary market” seen in the crypto primary market, but an important difference is that Claude has already formed real and rapidly growing enterprise revenue. Hence, what truly needs to be dissected is how much of this comes from actual commercialization versus the growth amplification effects resulting from continuous capital and resource inputs.
It is also important to pay attention to the statistical criteria for revenue. Anthropic includes some of the sales generated by Claude through cloud platforms at a more comprehensive amount in revenue, and then records the share paid to the partner platform as costs; some competitors might only recognize their final net revenue. Therefore, the annualized revenue of different AI companies cannot be directly compared horizontally. Media estimates suggest that if a more net-like criterion were used, Anthropic's revenue might decrease by about 6% to 10%. This won't negate its high-growth trend, but it will impact valuation comparisons. Anthropic is no longer just selling "the opportunity to chat with AI," but rather the capability to let AI accomplish actual work. As Claude begins to enter programming, internal enterprise processes, and software products, revenue increasingly depends on how much real work enterprises are willing to delegate to AI. This explains why Anthropic has become increasingly valuable, but exceeding $100 billion in annualized revenue merely answers the question of "is there anyone willing to pay." The more important question going forward is how much price the market is willing to pay in advance for this revenue.
3. Valuation of $2 trillion, exactly how much future has been bought in advance?
To understand the $2 trillion valuation, we can first look at the "valuation/annualized revenue multiple," which indicates how much the market is willing to pay for every $1 of the company's current annualized revenue rate. This metric helps to understand how high of a premium the market is paying for growth, but it cannot be equated to the traditional price-to-sales ratio calculated based on audited financial reports. In February, Anthropic completed a financing round at an estimated valuation of about $380 billion, with the annualized revenue at about $14 billion, corresponding to about 27.1 times annualized revenue. In May, the company's valuation rose to $965 billion, with annualized revenue exceeding $47 billion, corresponding to about 20.5 times. Based on the media report in September showing annualized revenue exceeding $100 billion, if the IPO valuation reaches $2 trillion, the corresponding multiple would be about 20 times or slightly lower. This set of changes provides an interesting perspective. Although Anthropic's valuation quickly rose from $380 billion to $965 billion and is further moving towards $2 trillion, the revenue growth rate is also rising in parallel. The $2 trillion valuation does not entirely stem from the market pushing the valuation multiple higher; it also results from investors continuously adjusting their estimates of Claude's revenue scale.
Figure 2: Anthropic's revenue and valuation acceleration timeline. Source: CoinW Research Institute
What truly determines whether the $2 trillion valuation can hold is whether future revenue can be realized. According to Reuters citing informed sources, Anthropic expects its revenue to reach about $190 billion to $200 billion in 2028. If this forecast is ultimately realized, the $2 trillion valuation would equate to approximately 10 to 10.5 times the 2028 revenue. In other words, the potential IPO valuation has already accounted in advance for Anthropic's continued near doubling growth over the next two years, positioning it to become one of the largest AI companies in the world in terms of revenue scale. This also means that the room for error left for the company is not wide. If revenue approaches $200 billion as expected while profit margins and cash flow continue to improve, the $2 trillion valuation could find long-term performance support; however, if revenue growth slows significantly, customers cut AI budgets, or computing power costs fail to decrease, the market may readjust the valuation multiples it is willing to pay. But revenue growth does not equal simultaneous value growth. What truly determines how much these revenues are ultimately worth is how much profit and cash flow Anthropic can generate from them. For Anthropic, the more difficult question to answer is how much computing power cost is required for every $1 of revenue generated by Claude.
4. Even with fast revenue growth, it must first outpace computing power bills
The biggest difference between Claude and traditional software lies in computing power. Once traditional software development is completed, the marginal cost of servicing an additional user typically does not significantly increase; however, each time Claude answers a question, generates a piece of code, or executes an Agent task, continuous operations of GPUs and data centers are needed behind it. The more users, the more powerful the model, and the more complex the tasks handled, the computing power demand will continue to rise. By 2025, this pressure had already become very evident. Anthropic's computing power and infrastructure expenses alone reached about $7.33 billion, surpassing that year’s revenue of $4.59 billion. However, this portion of expenses includes model training and may also include inference and infrastructure spending; it cannot be directly understood as the pure cost of servicing users, nor can it simply determine the gross profit margin of Claude's business.
What is more noteworthy is that Anthropic has already locked in massive computing power for future growth in advance. According to Reuters’ analysis of the confidential prospectus, the company’s future cloud computing, computing power, and infrastructure-related contract obligations total about $518 billion, with fulfillment periods generally extending to the next ten years. Of this, approximately 80% involves non-cancelable or minimum procurement and payment arrangements. This does not mean that Anthropic will spend $518 billion in one go in the short term, but it signifies that a significant portion of future costs has already been locked in through long-term contracts. The cost curve is not without improvement either. Media reports indicate that Anthropic achieved adjusted operating profit in positive territory in the second quarter of 2026, and expects to maintain positive figures in the third quarter. This indicates that revenue growth may have begun to outpace some operational costs. However, adjusted profit excludes items like stock-based compensation and cannot be directly equated to GAAP net profit or positive free cash flow. As hundreds of billions of dollars in long-term computing power contracts are gradually fulfilled, whether this profitability improvement can be sustained still requires subsequent financial reports for validation.
Thus, what Anthropic truly needs to prove is whether revenue can grow faster than computing power costs. If future inference unit costs decrease and model efficiency improves, the ratio of computing power spending to revenue will continue to decline, and true economies of scale will emerge; conversely, if each round of revenue expansion requires a nearly simultaneous increase in GPU and data center investment, Anthropic will struggle to fully enjoy the light-asset, high-margin business model of traditional software companies. There is also a potentially overlooked figure hidden behind high growth, which is the quality of revenue. In 2025, Anthropic's two largest clients contributed nearly a quarter of its revenue, and many large clients did not have long-term purchasing commitments. This not only implies a risk of customer concentration but also indicates that large clients and cloud platform partners may possess strong bargaining power. While current annualized revenue looks impressive, the market still needs to observe whether this revenue can truly stabilize into a steady, diversified, and sustained demand from enterprises. Therefore, what may warrant attention for Anthropic moving forward is not how many revenue records it has created, but rather how much computing power and cash investment are required for each $1 of revenue. When Claude can generate more revenue with less computing power, Anthropic will truly transition from being an AI company expanding based on huge capital inputs to a platform capable of continuously generating profits and cash flow.
Figure 3: Anthropic's growth flywheel and computing power bills. Source: CoinW Research Institute
5. The company is listed, but who does Claude ultimately listen to?
Apart from computing power, Anthropic has a valuation variable that is more unique than that of ordinary tech companies. Even with the company going public, public market investors may not necessarily hold the final decision-making power. According to Reuters citing the confidential prospectus, Anthropic's seven co-founders will form a Founder LLC and collectively control a special Class F share that holds 50.1% of the voting rights, including on certain key company matters such as some director elections and important matters for shareholder voting. Meanwhile, Anthropic will maintain its status as a public benefit corporation (PBC) in Delaware, with four other directors elected by the Long-Term Benefit Trust. Ordinary investors can buy Anthropic shares and share in the company’s economic growth, but it is difficult for them to influence its core direction as traditional public company shareholders might. Why would a company preparing to enter the public market retain such a structure? The answer relates to Anthropic's core AI safety principles.
The company even warns in its prospectus that more advanced AI might pose "catastrophic or existential risks," and states that as model capabilities and application scopes expand, related risks may also increase. Anthropic also acknowledges that safety research itself requires funding, talent, and computing power, and how much commercial return these investments can ultimately generate is uncertain. This creates an unusual contradiction for Anthropic. The market hopes for Claude to be stronger, to be used more, and for revenue to grow faster, but Anthropic's governance structure allows the founding team to prioritize safety and long-term missions over short-term commercial interests when necessary. The prospectus also alerts that such arrangements may lead to decisions that are not completely aligned with the company's financial interests. From a long-term perspective, this may not necessarily be a bad thing. Stricter safety boundaries may reduce serious incidents and regulatory risks, while possibly protecting Anthropic's brand and long-term value; but for the public market, it also means that Anthropic cannot be fully understood according to traditional tech company logic. Investors are not just purchasing Claude's future revenue and profits, but are accepting a governance mechanism where the founding team holds more control and may actively forgo certain business opportunities.
6. After reaching $2 trillion, what does the market really need to verify?
Anthropic has already proven that Claude has users, and that people are willing to pay increasingly more for it. After the IPO, what the market really needs to verify will no longer be just the growth rate, but the quality of that growth. First is revenue realization. Can the current annualized revenue exceeding $100 billion truly convert into quarterly and yearly revenue? Can the Claude Code, enterprise APIs, and Agent businesses continue to expand while reducing dependence on a few large clients? Since annualized revenue can be impacted by short-term growth and recognition criteria, the actual revenue reported in subsequent formal financial statements will be more important than a single run-rate figure. Secondly, computing power efficiency. As hundreds of billions of dollars in infrastructure contracts are gradually fulfilled, how much revenue can ultimately be generated from each $1 of computing power spending will determine whether Anthropic is closer to being a high-margin software platform or a heavily capital-dependent AI infrastructure company. Whether adjusted operating profit can remain positive and whether GAAP losses can narrow will also become important indicators for evaluating whether economies of scale have emerged.
Finally, cash flow and governance. Revenue growth ultimately needs to translate into self-sustaining capabilities, but for Anthropic, the issue is layered compared to ordinary tech companies. It not only needs to deliver performance but must also do so within AI safety boundaries it has set. The special governance structure of Founder LLC and Long-Term Benefit Trust is essentially designed to ensure that the company, while capital markets demand growth, retains control over model safety and long-term risks. This contradiction is becoming increasingly real. The recent meeting between Amodei and Trump has highlighted the most significant policy disparities currently facing the AI industry. Both parties agree that the U.S. needs to maintain AI leadership, as well as expand computing power and infrastructure, and have begun to form some consensus on internal model control and independent safety audits; however, they still hold differing opinions on whether the development pace of cutting-edge AI should be actively slowed. Trump is more concerned that regulation may slow the U.S. down, allowing China to catch up, while Amodei believes that too rapid advancements in model capabilities may lead to safety measures lagging, and continues to advocate for clearer safety standards and regulatory frameworks. AI safety for Anthropic is not merely a compliance cost; it is part of its business model. Stricter safety boundaries may increase testing, governance, and model development costs, even leading the company to voluntarily forgo certain business opportunities; but if the safety risks of cutting-edge AI are increasingly prioritized by government and enterprise clients, this mechanism may also turn out to be the foundation of trust for Anthropic to enter large enterprises, government, and high-sensitivity sectors. What really needs to be observed is whether this "safety premium" can long-term cover the growth constraints it brings.
Thus, the real bet that the $2 trillion valuation is placing is not whether Claude can continue to grow, but whether Anthropic can achieve the capital market's performance expectations that have been anticipated in advance under the premise of AI safety: keeping revenue on track toward the goal of approximately $200 billion by 2028; and ensuring that long-term revenue growth outpaces computing power costs, ultimately converting immense AI demand into profits and cash flow. The prospectus has allowed the market to see Claude's growth speed for the first time, but it has also made another matter clearer. Anthropic is not lacking in stories; what it truly needs to prove next is whether safety, growth, and profitability can coexist.
References
1.Reuters: Anthropic's IPO prospectus reveals financial data, losses, and computing costs
2.Reuters: Anthropic's path from AI startup to IPO
4.Reuters: Founder LLC, Class F shares, and founder control
5.Reuters: Anthropic warns AI may pose existential risks
6.Anthropic: Confidential draft of IPO application submitted
7.Anthropic: Completed $30 billion Series G financing in February 2026
8.Anthropic: Completed $65 billion Series H financing in May 2026
9.Axios: Anthropic's annualized revenue exceeds $100 billion
10.Axios: The difference in revenue recognition criteria between Anthropic and OpenAI
11.Anthropic: Expanding power cooperation with Amazon
12.The Irish Times cites FT: Anthropic's adjusted operating profit turns positive
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
