MiniMax breaks issue price, Zhipu pulls back: the gap between Chinese and US large models lies in market structure
MiniMax's stock price once fell below HK$160, down nearly 90% from its peak, while Zhipu fell more than 80% in four months. The author argues that the core gap between Chinese and US large models lies not in technology but in customer structure and willingness to pay. According to Rhodium Group data, from March to August 2026, the combined ARR of the model businesses of seven Chinese vendors—Alibaba, ByteDance, Kuaishou, DeepSeek, Moonshot AI, Zhipu AI and MiniMax—was about US$10.7 billion, while OpenAI and Anthropic combined exceeded US$100 billion over the same period; in terms of gross margin, Anthropic was about 65%, Zhipu about 26%, and MiniMax about 18%. MiniMax's total revenue in the first half of 2026 was about US$120 million, up 283.1% year on year, with B-end revenue up 703.1% and accounting for 63.4%, but its gross margin was 17.9%, down 11.8 percentage points from the fourth quarter of 2025, with price wars a major reason.
Why read this
AI score 71
Using financial reports and price-to-sales data from MiniMax and Zhipu, the article attributes the gap between Chinese and US large models to customer structure and willingness to pay, rather than model technology.
Topics
Body · AI translation
[[**MiniMax's stock broke below its IPO price and Zhipu suffered a sharp drawdown, exposing the core problem behind Chinese large models' revenue and gross margins falling far short of their US peers: the gap is not in technology but in customer structure and willingness to pay, and the price war will be hard to resolve in the short term.**
**Key points**
**1. Valuations remain divergent after the stock price crash**
MiniMax briefly fell below HK$160, down nearly 90% from its peak; Zhipu fell more than 80% in four months. As of October 22, 2026, their price-to-sales ratios were about 44x and 129x, while Anthropic's assumed valuation was about 76x.
**2. The revenue and gross margin gap between China and the US is stark**
From March to August 2026, the combined ARR of seven major Chinese AI vendors was about US$10.7 billion, while that of OpenAI and Anthropic exceeded US$100 billion; gross margin was about 65% for Anthropic, about 26% for Zhipu and about 18% for MiniMax.
**3. MiniMax's shift to B2B boosted revenue but depressed gross margin**
In the first half of 2026, MiniMax's total revenue was about US$120 million, up 283.1% year on year, with B2B revenue up 703.1% and accounting for 63.4%; however, gross margin was 17.9%, down 11.8 percentage points from the fourth quarter of 2025, with the price war a major reason.
**4. Market structure means the gap will be hard to narrow in the short term**
English text is being prepared.
It is probably hard for anyone to imagine that in just over half a year, MiniMax's stock price would briefly fall below HK$160, plunging nearly 90% from its high and marking its first break below the IPO price.
Zhipu, the first listed large-model company, is not much better off: in just four months, its stock price plunged more than 80%.
In half a year, the darling of the market turned into a has-been. What exactly happened behind this?
You might say: this is not a problem with Chinese AI, but rather that AI assets worldwide are being repriced. Take Anthropic: its original IPO valuation expectation of US$2 trillion has recently begun to be cut by some investors to US$1.5 trillion.
According to a Rhodium Group report, between March and August 2026, the combined annualized recurring revenue (ARR) of the model businesses of seven major Chinese AI developers—Alibaba, ByteDance, Kuaishou, DeepSeek, Moonshot AI, Zhipu AI and MiniMax—was about US$10.7 billion. Over the same period, the related revenue of OpenAI and Anthropic combined exceeded US$100 billion.
In terms of gross margin, Rhodium estimates that the overall gross margin of Chinese AI vendors is far below that of their US peers: about 65% for Anthropic, about 26% for Zhipu and only about 18% for MiniMax.
But what really matters is that the cause of the gap is not, at its core, technology—in fact, in many business scenarios, Chinese large models are already fully capable of substituting for top foreign large models.
The real gap comes from market structure.
Let us briefly break down MiniMax's 2026 interim report.
Total revenue was about US$120 million, up 283.1% year on year. Of this, consumer product revenue was about US$42.6 million, up 100.9% year on year, with its share falling to 36.6%. B2B product revenue was about US$73.9 million, surging 703.1% year on year, with its share of total revenue jumping from 30.3% to 63.4%, making it the largest source of revenue.
In other words, MiniMax is no longer the "consumer AI company" it was when it first listed; it has become an AI company mainly focused on B2B products.
In addition, according to the financial report, B2B products contributed about 75% of MiniMax's growth, so we can boldly infer that MiniMax's B2B revenue share will continue to surge, eventually exceeding 90%, making MiniMax thoroughly a B2B AI company.
However, MiniMax's market transformation has brought not only revenue growth but also a further decline in gross margin.
According to the financial report data, MiniMax's gross margin in the first half of 2026 was about 17.9%, down about 11.8 percentage points from 29.7% in the fourth quarter of 2025.
One of the core reasons for the decline in gross margin is the price war.
For example, in early June, when MiniMax's latest model M3 went online, MiniMax switched its Coding Plan, originally billed by number of requests, to a Token Plan billed by actual consumption, which at one point sparked user complaints about a price increase. To placate users, MiniMax had to announce that the M3 API price would be permanently halved and set up different tiers of benefits or bonus quotas for new and existing users.
You might say that MiniMax's model capabilities are inferior to Zhipu's, so it can only cut prices to promote sales. But look at Zhipu's gross margin in the first half of 2026: 26.4%, which in fact is not far from MiniMax's.
So the real question is: why can Chinese large models only compete through price wars?
Some will say: this is because the capabilities of Chinese large models still fall short of Claude's top models, but this logic only applies to overseas markets.
In China, most enterprises can only use domestic large models. In a market where the strongest enemy has already been blocked, why can we still only win through price?
Fierce competition is certainly one factor, but that alone is not enough to explain the huge gross margin gap between the Chinese and US markets; after all, Anthropic likewise faces competition from OpenAI as well as giants such as Google and Meta.
I think the core reason is still market structure.
However, for traditional enterprises such as manufacturing and wholesale trade, AI is only an auxiliary tool. Although it can improve the efficiency of some administrative work, such as writing weekly reports, organizing data, or even entering ERP documents, its help to the core business is very limited.
In Europe and the United States, there are a large number of technology enterprises in software, financial services, and other fields, and they are the foundation of the high gross margins of US large models. In China, by contrast, traditional enterprises such as manufacturing and wholesale trade are the mainstream.
It is not only the corporate market structure; the gap in profit levels between Chinese and US enterprises is also very large.
Take manufacturing as an example. According to the "White Paper on a Comparative Study of China-US Fortune 500 Enterprises" released by the CCID Research Institute in December 2020, since 2016, the average profit of US manufacturing enterprises on the list has been about 4.9 times that of Chinese manufacturing enterprises on the list.
Now look at China's software industry. A 2024 report by the CCID Research Institute shows that the overall profit of Chinese listed software companies has declined for three consecutive years, and in the first half of 2024 it fell 91.62% year on year. Although revenue reached as high as 298.8 billion yuan, net profit was only 690 million yuan, earning just 0.2 fen per 1 yuan.
Of course, you may say that these are old data and do not represent the recent situation. But frankly speaking, in recent years has the profit level of most Chinese manufacturing enterprises risen or fallen? Has the profit level of software companies risen or fallen? I think the answer is not so optimistic.
Of course, a gap is not frightening. We can also quickly catch up in technological gaps. But what is truly frightening is the gap in market structure, because in the short term it is an unsolvable problem.
More importantly, products and markets reinforce each other: enterprises with greater willingness and ability to pay will prompt AI companies to further increase investment, thereby producing better AI products; and better products will bring greater value to enterprises and boost their willingness to buy.
From this perspective, although the revenue of US AI companies is already far ahead of ours, in the coming years the revenue gap between the Chinese and US AI industries will not narrow but will further widen. Correspondingly, the valuation gap between Chinese and US AI companies will also further widen.
But what is the actual situation?
We can compare the valuations of Chinese and US large-model companies.
(Note: The following measures are only for convenience of comparison, not professional financial measures, and should not be used for investment reference!)
To unify the measures, we first use revenue for the 12 months ended the end of June 2026 to look back at the ratio of market value to revenue (price-to-sales ratio) of Zhipu and MiniMax at their stock price highs and at their lows.
When Zhipu's stock price reached a high of HK$2,980 on June 22, 2026, its market value was about HK$1.3 trillion. Based on revenue of about HK$1.736 billion for the 12 months ended the end of June 2026, the corresponding ratio of market value to revenue (price-to-sales ratio) was about 749 times.
When MiniMax's stock price reached a high of HK$1,330 on March 18, 2026, its market value was about HK$410 billion. Based on revenue of about HK$1.295 billion for the same revenue period, the corresponding ratio of market value to revenue (price-to-sales ratio) was about 317 times.
After the stock prices continued to fall, the price-to-sales ratios of the two companies have clearly declined.
As of October 22, 2026, Zhipu's market value was about HK$224.7 billion, and based on revenue for the 12 months ended the end of June 2026, its price-to-sales ratio was about 129 times; MiniMax's market value was about HK$56.6 billion, and its price-to-sales ratio was about 44 times.
What about Anthropic?
Anthropic has not yet gone public. If calculated according to the US$1.5 trillion valuation discussed by some investors and the revenue for the same period reported by the media, the ratio of its valuation to revenue for the most recent 12 months (price-to-sales ratio) is about 76 times.
That is to say, after a substantial correction, under a similar comparison basis, MiniMax's price-to-sales ratio (44 times) is already lower than Anthropic's assumed valuation multiple (76 times), while Zhipu's valuation multiple (129 times) is still higher than Anthropic's.
Of course, I am not bearish on China's AI industry. In fact, despite many difficulties, most software companies are actively pursuing AI transformation. At the same time, through co-creation with customers, many excellent enterprise AI applications have already emerged.
The prospects of AI are certainly bright, but we still have a lot of hard work to do. AI industry signal channel: Frontier Technology. This content comes from the internet and the original link. The views represent only the author and do not represent the position of Huxiu. If there is any copyright issue, please contact hezuo@huxiu.com, and we will verify and handle it promptly. Everyone who is changing the world and wants to change the world is in the Huxiu APP[]