Don’t Only Ask About Profitability

When asking “Do large model companies have positive profits?”, one must first ask which layer or level of profitability is being referred to.

A positive gross profit only indicates that after products and services are sold, the direct costs have not completely consumed the revenue. A narrowing of adjusted losses suggests that the loss rate may improve after excluding certain non-cash or one-time items. Only when net profit turns positive and operating cash flow turns positive is it closer to what ordinary people understand as “the company can sustain itself.”

Large AI model companies are most likely to get stuck here. Inference costs can decrease through engineering optimization, and APIs, enterprise services, and subscription products can all generate revenue. However, as long as frontier models continue high-frequency training, the compute power, cloud services, chips, human capital, and data engineering will once again suppress the profit and loss statement.

Therefore, this article draws one limited conclusion: As of early June 2026, public financial materials show that Zhipu and MiniMax, two representative independent large model companies, are still incurring losses; while Anthropic and OpenAI’s public materials demonstrate significant fundraising and revenue scale, they do not yet prove that they have reached the profitability inflection point in terms of net profit.

Domestic Ledgers are Very Direct

Zhipu’s revenue in 2025 was RMB 7.243 billion, with a gross profit of RMB 2.967 billion. This means it is no longer in a state of “lack of commercialization.” The problem is that the same HKEX annual performance announcement reports R&D expenditure of RMB 31.804 billion, an annual loss of RMB 47.182 billion, and an adjusted net loss for the year of RMB 31.820 billion.

This proportion is alarming. The issue isn’t that revenue hasn’t increased; rather, the money spent by the company to maintain model capabilities and fuel commercial expansion far exceeds the annual income.

MiniMax’s books are similar, only the currency is changed to USD. In 2025, revenue was $790.38 million, gross profit was $200.79 million, and the gross margin increased from 12.2% in 2024 to 25.4%. This indicates that the model and system efficiency are improving, and that products are indeed being paid for by users. However, the company incurred an annual loss of $1.8716 billion, an adjusted net loss of $0.2509 billion, and had R&D expenditures of $0.2528 billion.

Neither of these two companies have zero revenue, nor do they have zero gross profit. To be more accurate: revenue growth has begun, and gross profit is improving, but the fixed investment in R&D and model competition has not yet been covered by commercial revenue.

Return A is the externalization of funding needs

After listing on the Hong Kong Stock Exchange (HKEX) in January this year, MiniMax also initiated A-share IPO preparation at the end of May. Xinhua Finance reported that the company filed an A-share IPO guidance filing on May 29, with CITIC Securities serving as the guiding institution; MiniMax subsequently stated in an HKEX announcement that its Board has resolved to explore preliminary suggestions for issuing RMB shares, is evaluating listing on the STAR Market, and has signed a guidance agreement.

Zhipu’s progress is becoming more concrete. Announcements and circulars disclosed by the Hong Kong Stock Exchange (HKEX) in early June show that the company plans to propose, at its Annual General Meeting, an A-share issuance and listing on the STAR Market. Details such as the issue size, target investors, issuance methods, and fundraising projects have all been included in the draft resolution text. The circular also clearly states that the issuance and listing still require approval from the Shareholders’ Meeting, regulatory authorities, and the Shanghai Stock Exchange (SSE), meaning that the A-share listing has not yet been completed.

The issue should not be framed as merely a single narrative of “going public due to lack of funds.” The A+H dual platform can provide local currency financing, valuation anchors, institutional coverage, and liquidity for employees and early investors, which also strengthens the company’s capital market standing when dealing with domestic industry chains and government/enterprise clients.

But the rhythm itself still raises questions. Hong Kong stocks just finished pricing, and A shares quickly followed suit, which is unlike traditional software companies that slowly expand their capital platforms after stabilizing their profit models. Large model companies require a more continuous access to capital because they are simultaneously spending money on model generations, inference scale, client delivery, and global channels.

In other words, [Action referring to ‘return to A’] can be seen as a strategic move, but it also represents the externalization of capital needs. The capital market is not adding superficial enhancements to companies that are already highly profitable; rather, it is fueling an infrastructure race that has not stopped.

Two American Companies Couldn’t Provide a Complete Answer

Anthropic’s moves are closer to the standard U.S. IPO process. The company announced on June 1, 2026, that it had made a confidential submission of the Form S-1 draft registration statement to the SEC. Here, “confidential” does not mean conspiratorial secrecy, but rather a confidential submission. Public investors cannot see the complete prospectus yet, nor can they see the number of shares, issue price, or audited profit statements.

Its pace of commercialization is truly astonishing. On May 28th, Anthropic announced a Series H funding round of $65 billion, with a post-money valuation of $965 billion, and stated that its run-rate revenue already exceeds $47 billion. However, run-rate revenue annualizes the current revenue speed; it does not equal full-year revenue, nor does it equate to net profit.

OpenAI faces a similar situation. Officially, at the end of March, it announced the completion of $122 billion in committed capital financing, with an investment-after valuation of $852 billion. It also stated that current monthly revenue is $2 billion, and corporate revenue accounts for over 40%. These figures are sufficient to demonstrate that its commercial scale has far surpassed the early laboratory phase.

However, OpenAI’s IPO timeline is not officially confirmed. Axios reported on May 20th that OpenAI is preparing a confidential IPO prospectus, which might be submitted soon, but the timing could still change. Furthermore, the response from an OpenAI spokesperson merely stated that the company would evaluate strategic options through normal governance and that its focus remains on execution. It would be too much to directly rephrase this news as “OpenAI listing in the second half of the year.”

For Anthropic and OpenAI, what we truly await are their public S-1 filings and complete financial disclosures. At that point, the focus should not be on how exaggerated the valuation is, but rather on metrics such as gross margin, R&D expenses, cloud and chip contracts, customer concentration, deferred revenue, CapEx commitments, and whether losses are narrowing relative to revenue growth.

Capital Window Precedes Profit Window

The unique situation for large model companies is that they no longer merely talk about technological vision like they did in 2023. Revenue is real, paying customers are real, developer usage is real, and enterprise integration is also happening.

However, they are not like traditional SaaS, where profits naturally rise after customer acquisition costs decline, renewal rates stabilize, and gross margins expand. The competition among frontier models continuously resets the cost base: As soon as the previous generation of models lowers inference costs, the next generation—with its requirements for training, long context handling, multimodality, agent execution, and infrastructure redundancy—will raise the overall investment level again.

This is also why the listing window appears earlier than the profitability window. The capital market looks at a company that has the potential to become an infrastructural layer, not just a clean profit statement. Companies are looking for larger-scale, longer-term, cheaper, or more diversified sources of funding, rather than short-term proof that they can reduce expenses.

This does not mean that large model companies will necessarily fail to achieve profitability. A more reasonable observation is that they are transitioning from having “zero revenue” to “rapidly growing revenue,” but have not yet universally reached a point where “commercial revenue is sufficient for self-sustainment.” In this intermediate phase, fundraising is not an ancillary action, but rather an integrated part of the business model.

Therefore, the general direction in the prompt is correct. However, the conclusion needs to be refined: not all large model companies are still stuck in a pre-revenue phase, nor can we say that every company lacks positive gross profit; but if viewed through stricter metrics like net profit and operating cash flow, overall large model companies have not yet escaped capital dependency. What they are currently vying for is not a mature “harvesting window,” but the sustainable edge (or operational longevity) in the next round of model competition.

References

Author’s Note / Appendix Notes

Original Prompt

$blog-writer AI large model companies, none of them seem to have started generating positive profits yet. Model R&D is expensive, and both Zhipu and Minimax are considering returning to A-shares for a secondary listing and round of financing. Claude’s parent company has secretly submitted its prospectus, and OpenAI also plans to list in the second half of the year.

Writing Outline Summary

This article retains several capital market movements mentioned in the prompt: Zhipu and MiniMax returning to A funding, Anthropic filing S-1, and reports of OpenAI preparing for an IPO. We eliminated two potentially misleading statements: first, stating that all large model companies have “no positive profitability,” and second, presenting OpenAI’s media news as a confirmed listing plan.

The main body breaks down “profitability” into several layers—namely gross profit, adjusted loss, net income, and cash flow—and draws conclusions only where public materials support them. It did not elaborate on stock prices, market capitalization, or investment recommendations because doing so would shift the focus to trading judgment, thereby diluting the core accounting issues this piece aims to explain.