Looking at Zhipu and MiniMax's Growth, Valuation, and Long-Term Narrative Through Anthropic's Prospectus

On October 26, Anthropic publicly filed its prospectus with the U.S. Securities and Exchange Commission, planning to list on Nasdaq under the ticker "ANT." For the first time, its financial data, customer structure, compute contracts, and governance arrangements as of June 30, 2026, were fully laid before the market.

On the financial figures, Anthropic's revenue in the first half of 2026 was $16.286 billion, up about 1,300% year over year; its operating loss was $693 million, adjusted operating profit was $1.618 billion; and cash flow from operating activities was positive at $3.929 billion. At the same time, the reported net loss reached $80.748 billion, but about $80 billion of that was caused by non-operating items, mainly fair value changes in financial instruments such as equity and convertible bonds, investment losses, and related one-time accounting adjustments.

On costs and expenses, Anthropic's compute and infrastructure expenses in the first half were $8.945 billion, accounting for 54.9% of revenue; as of the end of June it held $112.389 billion in cash and $19.262 billion in short-term investments, but financing cash flow in the same period also reached $110.774 billion. On compute leasing, the company's main non-cancelable compute contracts with the other three major cloud providers, Google, Amazon, and Microsoft, totaled about $269.4 billion.

Rapid growth, positive operating cash flow, and long-term compute commitments exist simultaneously, constituting Anthropic's current fundamentals.

"Bright Company" previously mentioned that, as SpaceX goes public and Anthropic and OpenAI successively enter the listing stage, the significance of their public financial data also lies in giving the market a clearer "pricing anchor." From the financial to the narrative level, Anthropic will also become a reference for Zhipu (02513.HK) and MiniMax (00100.HK), as well as DeepSeek and Moonshot AI, which may list in the future.

Interestingly, on the first trading day after the prospectus was published (the 27th), as of press time, the share prices of Zhipu and MiniMax both rose, but Zhipu's gain (7.81%) was larger, which also reflects market preference to some extent.

Anthropic: A company with unprecedented growth

Judging from the prospectus, Anthropic's first impression is that of a company with "unprecedented" growth. The company's revenue grew from $41 million in 2023 to $386 million in 2024, and reached $4.585 billion in 2025. Revenue in the first half of 2026 was $16.286 billion, compared with $1.163 billion in the same period last year, up about 1,300% year over year. Half-year revenue was already 3.6 times that of all of 2025.

By comparison, the company's losses are also narrowing sharply, and operating cash flow has already turned positive. In the first half of 2026, Anthropic's operating loss was $693 million, an operating loss margin of about 4.3%, compared with a loss of $2.135 billion in the same period last year. Adjusted operating profit was $1.618 billion, and adjusted net profit was $2.671 billion. Cash flow from operating activities was positive at $3.929 billion.

As for the loss, the $80.748 billion net loss in its prospectus needs to be broken down specifically: $72.388 billion came from fair value changes in convertible bonds, derivative instruments, and SAFEs, and another $8.757 billion was net other income and expenses, mainly also related to revaluation of financial instruments. It should be noted here that a rise in Anthropic's valuation will increase the carrying value of related liabilities, thus generating huge non-cash expenses; the net loss is more affected by the accounting treatment of financing instruments and does not directly affect cash flow.

Compared with traditional companies, Anthropic did not provide a gross margin. Starting in 2025, the company listed expenses by nature rather than dividing them into cost of revenue, R&D, sales, and administrative expenses. Among them, in the first half of 2026, compute and infrastructure expenses were $8.945 billion, accounting for 54.9% of revenue. In the same period, the company's personnel expenses were $3.152 billion, sales, marketing, and channel expenses were $3.168 billion, and other operating and administrative expenses were $1.714 billion. Total operating expenses were $16.979 billion, slightly higher than revenue.

Perhaps a more meaningful signal here is that compute expenses as a share of revenue fell from 201% in the first half of 2025 to 55%, meaning revenue grew faster than the already locked-in compute base.

Anthropic's revenue structure is also different from that of traditional software companies. In the first half of 2026, usage-based billing revenue accounted for 80%, and subscription revenue accounted for 20%. Cloud marketplace channels contributed 48% of revenue; in terms of payment, the company contracts with end customers and then pays channel fees to AWS, Google Cloud, and Microsoft Azure.

On customers, as of the end of September, the company had 11,900 customers with annual spending of more than $100,000, and a net revenue retention rate of about 900%. U.S. investor Greg Isenberg believes this is equivalent to "customers spending about 9 times more after one year of use, compared with excellent SaaS companies that typically treat a 120% net revenue retention rate as a bragging point." However, this metric also seems unsuitable for linear extrapolation over the long term.

Anthropic did not disclose specific ARR figures. In versions previously circulating in the market, Primary's figure was $67.2 billion, and NYT, citing people familiar with the matter, reported year-end ARR guidance that could reach $100 billion; most mainstream media put the number between $65 billion and $67 billion.

Comparing Zhipu and MiniMax financial figures: growth rates are already the most similar part

If these figures are compared with those of Zhipu and MiniMax, the two Chinese AI large-model companies already listed, many different logics can be found.

For the first half of 2026, Anthropic's revenue was $16.286 billion, up about 1,300% year on year; operating loss was $693 million, adjusted operating profit was $1.618 billion, and net operating cash inflow was $3.929 billion.

Zhipu's revenue was RMB 954 million, up 399.7% year on year; gross margin was 26.4%, adjusted net loss was RMB 1.964 billion, and net operating cash outflow was RMB 3.374 billion.

MiniMax's revenue was $117 million, up 283.1% year on year; gross margin was 17.9%, adjusted net loss was $293 million, and net operating cash outflow was $318 million.

Source: company prospectuses, financial reports

From the above data, Anthropic has already passed the inflection point in operating cash flow, while Zhipu and MiniMax are still in the "cash-burning stage."

The revenue growth rates and cash flow status of the three companies in the first half of 2026 are also reflected in revenue and spending.

As mentioned earlier, the reported gross margins of Zhipu and MiniMax can be compared with each other, but cannot be directly compared with Anthropic. Zhipu and MiniMax put model training investment in R&D expenses, while cost of sales mainly reflects inference and delivery. Anthropic discloses training, inference and internal compute together.

If comparing investment across the entire operating system, Zhipu and MiniMax are still in a high-investment stage. In the first half, R&D expenses were equivalent to 223% and 255% of revenue, respectively.

On the revenue side, the revenue structures of the three companies are converging, but their starting points differ.

In terms of revenue model, Anthropic's usage-based billing revenue accounts for 80%, and subscriptions and other revenue account for 20%; in terms of channels, cloud marketplace channels contribute 48%, direct sales and partner channels contribute 32%, and the core is enterprise customers continuously purchasing tokens through APIs and cloud channels.

Zhipu previously focused on government and enterprise customers and localized deployment. At least in terms of revenue model, Zhipu's proportions are close to Anthropic's. In the first half of 2026, open platform and API revenue rose to RMB 825 million, accounting for 86.5% of total revenue, while the share of localized deployment revenue fell to 13.5%. The company shifted from selling projects to selling usage volume, and revenue began to be directly tied to token consumption.

MiniMax previously leaned more toward consumer AI products. In 2025, AI-native products contributed about two-thirds of revenue. By the first half of 2026, open platform and enterprise services revenue grew 703.1% year on year, and its share rose to 63.4%; AI-native products accounted for 36.6%.

The company still retains video, voice, music and AI companion products, while making enterprise and developer platforms a new growth source. Overseas revenue accounts for 60.8%, which is MiniMax's clearest difference.

Combining the previous data, we can further compare the valuation levels of Zhipu and MiniMax.

As of the close on October 26, MiniMax's market value was about HK$61.4 billion, equivalent to about $7.9 billion; Zhipu's market value was about HK$235.4 billion, equivalent to about $30.2 billion.

Considering that Anthropic is not yet listed, the latest round of about $965 billion in financing is still a primary-market valuation, while secondary-market valuation expectations may be between $1.5 trillion and $2 trillion. Among market voices, those close to a $1.5 trillion valuation seem to be more numerous.

Now look at ARR. Public operating data say MiniMax's ARR exceeded $800 million in August, and Zhipu's ARR was about $1.6 billion at the end of August.

Based on the above market values, MiniMax's P/ARR is about 9.8 times, and Zhipu's about 18.9 times. Anthropic's latest financing round valued it at about $965 billion; using ARR of about $67.2 billion, that corresponds to about 14.4 times; based on the $1.5 trillion IPO valuation discussed in the market, it would be about 22 times.

Source: AlphaEngine compilation

Therefore, it is not that "Anthropic is cheap and the two Hong Kong-listed companies are expensive," but rather that a valuation hierarchy has formed. MiniMax has the lowest multiple, Anthropic's private valuation is in the middle, and Zhipu is higher; if Anthropic lists at a $1.5 trillion valuation, it would again become the highest-multiple company among the three.

Under the static assumption that ARR remains unchanged at about $67.2 billion, its P/ARR is about 22 to 29.8 times; if ARR continues to grow, the actual multiple will be lower than this level.

This range means that Anthropic needs to simultaneously prove revenue growth, gross margin improvement, declines in model training and inference costs, and the sustainability of enterprise customer contracts; otherwise, $2 trillion looks more like an advance draw on revenue and profit for years to come.

From a relative valuation perspective, once Anthropic reaches $1.5 trillion, its P/ARR will be higher than those of Zhipu and MiniMax; at $2 trillion, the valuation premium will become even more pronounced. It may become the pricing anchor for global large-model companies, but it will also raise the valuation threshold for the entire sector: if Anthropic can list at a higher multiple and continue to deliver growth, Hong Kong-listed model companies may receive a valuation re-rating; if it pulls back after listing due to high valuations, the relative discounts of Zhipu and MiniMax may not be enough to offset industry risk.

Source: AlphaEngine compilation

If we further annualize first-half revenue, the difference is even greater. Anthropic is about 30 times at a $965 billion valuation and about 46 times at a $1.5 trillion valuation. If it rises further to $2 trillion, that corresponds to about 61 times annualized first-half revenue. MiniMax is about 34 times. Zhipu is about more than 100 times on the company-disclosed USD-converted basis.

Zhipu's ARR is far higher than its annualized first-half revenue, indicating that its growth mainly occurred after the end of the reporting period, and also that its valuation relies more on future revenue delivery.

There are several reasons Zhipu commands a higher multiple. It is one of the few pure foundation-model companies in Hong Kong stocks, with labels such as domestic models, domestic computing power adaptation, and government and enterprise customers. The company has also shifted its business logic from local deployment to API, Coding, and agents, and the market is willing to pay for changes in the revenue structure.

Long-term narrative: inside and outside technology

If you read the "Letter to Investors" at the beginning of Anthropic's prospectus, you may find that it describes the relationship between AI and the company from the height of a "civilization"—even more as if the company is a kind of "sense of mission" and "chosen one" bringing AI to the world. In particular, the company talks about technology's impact on society and believes that Scaling Laws will still drive the growth of machine cognitive capabilities, ultimately forming a "country of geniuses in the data center." Millions of agents will enter software, science, medicine, finance, government, and national security. AI will not only improve labor efficiency but also change the innovation process itself.

Compared with Zhipu and MiniMax, Anthropic particularly emphasizes the industry potential of AI in biomedicine. In the letter, it writes: "First, we believe that AI will greatly accelerate progress in biology and medicine, potentially curing diseases that have plagued humanity for thousands of years, including cancer, Alzheimer's disease, heart disease, rare genetic diseases, and major infectious diseases. AI is creating entirely new ways of scientific research and drug development. We want to play a leading role in developing tools that serve the broader scientific community. Anthropic has already partnered with many biotechnology and pharmaceutical companies to improve basic research and clinical trials. From helping pharmaceutical company scientists solve difficult problems in drug discovery to shortening the writing time of clinical trial reports from months to minutes, we are involved."

The letter also places risk within the vision. Anthropic divides risk into misuse, loss of control, authoritarian exploitation, and economic shock. The company explicitly states that it may delay model releases, give up short-term revenue, and even proactively keep its pace of development below what is technically achievable. Moreover, Claude will remain ad-free. The company may also make decisions that do not conform to the principle of profit maximization.

Moreover, these commitments are linked to corporate governance. Anthropic is a public benefit corporation, and the Long-Term Benefit Trust can elect a majority of the board. The company has established a public interest fund with an initial contribution of 5% equity; when the valuation reaches certain milestones, the board can continue to contribute equity.

By contrast, Zhipu's narrative is closer to a technology roadmap. The prospectus proposes "making machines think like humans and benefiting humanity with reliable AI," and divides AGI evolution into multiple stages. The latest interim report further proposes a capability ladder: Chat, Coding, Agent, Co-work, and Autonomous AI. Models first answer questions, then complete code, then enter professional workflows, and ultimately undertake long-term goals under unattended conditions.

Zhipu writes this route as a business formula: the upper bound of intelligence expands task boundaries, and task value determines the economic value that tokens can carry. The company focuses on long-horizon tasks, reinforcement learning, task environments, and recursive self-improvement. Domestic chip clusters, the MaaS platform, and the open-source ecosystem form its infrastructure narrative.

Security is also defined as a verifiable public good, building trust through red-team testing, vulnerability disclosure, external evaluation, and controlled open access to sensitive capabilities.

MiniMax's mission in its prospectus is "Intelligence with Everyone." The company emphasizes multimodality, inference efficiency, and global accessibility. The goal is for models to handle programming and complex agent tasks, generate video, voice, and music, and enter the daily lives of ordinary users. MiniMax's core statement is the "performance-cost frontier." Model capabilities must first meet task requirements, and then inference costs are reduced through joint optimization of models, systems, infrastructure, and products. Only after costs fall do more tasks have commercial value and more users continue to call the models.

In its prospectus, security is mainly expressed as data filtering, alignment, content governance, and human oversight, and has not become an investor-facing governance contract.

Overall, Anthropic is the "developer and governor of AI," Zhipu is the "builder of AGI capabilities and infrastructure," and MiniMax is the "platform for global AI products and AI inclusiveness." All three believe that model capabilities will continue to improve, but each has chosen a different way to prove it.

In summary, Anthropic's listing will make the market focus more on revenue, gross margin, computing costs, and cash flow. MiniMax's advantage lies in globalization and product diversification, while Zhipu benefits from China's government and enterprise market, open-source ecosystem, and localization experience. Clearly, both companies need to prove that ARR can be converted into real revenue and improve gross margin and operating cash flow.

This article comes from the WeChat public account "Bright Company" (ID: suchbright), author: Editor-in-Chief online 24 hours, published by 36Kr with authorization.