Over the past six months, commercialization of domestic large language models has suddenly taken off.
In December 2025, the annualized revenue of China's major AI model companies was only $4 billion, but now it has approached $13 billion. Growth has accelerated notably especially since June.
What's more counterintuitive is that this growth wasn't achieved through price wars.
Zhipu's average API prices rose 101% this year, yet token usage grew more than 40-fold. DeepSeek also raised prices significantly, but usage didn't drop noticeably.
Meanwhile, overseas expansion has become an important growth driver. MiniMax generates over 60% of its revenue from overseas, Kimi's overseas revenue has already surpassed its domestic revenue, and Zhipu's overseas share is estimated at nearly 40%. This marks the first time Chinese AI has achieved software exports at a meaningful scale.
While revenue grows, valuations of model companies are diverging. Based on the latest ARR, DeepSeek trades at less than 70 times ARR, while Zhipu and MiniMax are at 36 times and 20 times, respectively.
Why has domestic large model revenue suddenly accelerated? And why do valuations differ so much among model companies? Today, let's discuss these two questions.
No Chinese text was provided for translation.
Chinese AI model revenue growth is accelerating.
Data show growth in China's large models has been accelerating since June.
In December 2025, China's major AI model companies had annualized revenue of about $4 billion. By March this year, that had risen to $6 billion. Now it is close to $13 billion.
In other words, in eight months, the annualized revenue of China's AI model companies has climbed from $4 billion to $13 billion.
And the pace is still accelerating.
From December 2025 to March this year, annualized revenue rose by about $2 billion in three months. From March to June, it added another $4 billion. From June to August, it added about $3 billion in just two months.
In other words, ARR was initially adding about $600 million to $700 million per month. Over the past two months, monthly additions have exceeded $1 billion. Commercialization of large models is clearly picking up speed.
The most representative example is Zhipu AI. In March, Zhipu's MaaS — its API business — had ARR of only about $250 million. By July, it was close to $1 billion. In August, annualizing monthly revenue put ARR at $1.6 billion. Annualizing the latest week's revenue, it has even reached $2 billion.
August API revenue alone surpassed Zhipu's total API revenue for the first half of 2026.
If current trends hold, Zhipu's API ARR could approach $2.5 billion by year-end.
MiniMax shows a similar trajectory. Its ARR stood at roughly $100 million in December 2025, surpassed $150 million in February 2026, and reached $300 million to $430 million by April-May. By August, annualizing the latest weekly revenue put ARR above $800 million.
Management previously guided to revenue exceeding $1 billion by the end of 2026.
Two counterintuitive trends underlie this growth:
First, growth hasn't come from price cuts.
Price wars have been a hallmark of China's large-model industry over the past two years, with models and tokens getting cheaper. So the initial reaction to rising usage and revenue might be that companies are again sacrificing price for volume.
But the latest data suggest the opposite.
Starting in August, API prices for various DeepSeek products increased by roughly 3 to 12 times.
Based on peak-hour rates, compared with the old prices at the end of July, different billing items generally rose by about 3 to 12 times.
Among them, the V4 Pro cache-hit input price rose from $0.003625 to $0.044 per million tokens, an increase of more than 12 times; the output price rose from $0.87 to $3.96, up about 4.6 times.
Despite such large price increases, usage has not dropped significantly.
Zhipu is similar. Since the start of the year, its average API prices have risen 101%, roughly doubling. At the same time, token usage has grown more than 40-fold.
The second change is overseas revenue.
According to estimates by FD, author of Robonomics, overseas revenue of Chinese AI model companies has now exceeded $2 billion.
Among them, MiniMax has the highest share of overseas revenue, with 60.8% of its revenue coming from abroad.
Kimi also generates a large share of revenue overseas. In February, reports said its international revenue surpassed domestic for the first time. By June, annualized recurring revenue had grown to about $300 million, with overseas developers seen as a key growth driver.
The author estimates Zhipu's overseas revenue share has also reached roughly 40%. Even giants like Alibaba and Tencent, whose revenue is mostly domestic, are estimated to derive 5% to 10% of model revenue from overseas.
This marks the first time Chinese AI has produced software exports at meaningful scale.
It looks like your message may have been cut off. Could you paste the full Chinese article you'd like translated?
Valuations are diverging, ranging from 20 to 70 times ARR.
Even as revenue growth accelerates, a new problem is emerging:
Large-model companies are becoming harder to value.
For now, valuations of domestic large models have clearly diverged.
DeepSeek is valued at around $74 billion based on the latest funding-round chatter, with annualized recurring revenue exceeding $1 billion, implying a multiple of under 70x. Zhipu has a market valuation of about $71 billion and latest weekly ARR of roughly $2 billion, implying 36x. MiniMax is valued at about $15.6 billion with ARR of approximately $800 million, implying around 20x.
From 20x to 70x is a gap of more than threefold, far exceeding the valuation spread among overseas foundation-model companies. By comparison, since 2H25, the valuation gap between OpenAI and Anthropic has remained around 1x.
Why is the valuation gap among Chinese foundation-model companies so wide?
One important reason is that strategic divergence among Chinese players is greater than that between OpenAI and Anthropic.
Both OpenAI and Anthropic have top-tier model capabilities and ample compute, capital and R&D resources.
Although their early areas of strength were not exactly the same, once a direction is validated, both have the ability to quickly reallocate resources.
The most typical example is coding. After coding proved to be one of AI's most important high-value use cases, OpenAI quickly began concentrating resources to catch up.
Third-party tracker TickerTrends estimates that Codex's annualized revenue was only about $1 billion in February this year, rose to $5.6 billion by June and reached $8.8 billion by August. Over the same period, Claude Code generated about $15.1 billion in annualized revenue.
Claude Code remains ahead, but the revenue gap between the two has narrowed from roughly 2.4 times in early July to 1.7 times in August.
Domestic players, however, are a different story entirely.
Given limited resources, no one can pursue every frontier at once, so each company must place early bets on a limited set of directions. For example, Zhipu early on leaned toward MaaS, enterprise APIs, and domestic compute adaptation; Kimi bet on long-context; MiniMax committed resources earlier to multimodal, video, and overseas markets.
Unlike OpenAI, Chinese model companies can't simply pour billions into catching up once a direction proves wrong. So when consensus shifts on the right direction, their strategic adjustments often take months or longer.
As a result, Chinese model companies' business outcomes depend more on early strategic judgment than those of leading overseas firms. The greater the strategic divergence, the wider the revenue gap, and the harder valuations become to converge.
Beyond strategic differentiation, companies also differ in efficiency.
The Information earlier reported that DeepSeek's API business gross margin reached 82.9% in the first seven months of this year, while Anthropic's is estimated at over 60%. That puts DeepSeek's gross margin above Anthropic's.
By contrast, Zhipu's API gross margin is just 24.6%, and most domestic text-model APIs still have gross margins in the 40%-60% range.
The gap is striking. At $1 billion ARR, an 80% gross margin leaves $800 million in gross profit, while a 20% margin leaves just $200 million — a fourfold difference.
Combined, these factors make it hard to value Chinese model companies with a single uniform ARR multiple.
What truly separates valuations are two things: growth trajectory and revenue quality.
The former determines how fast a company can scale; the latter determines how valuable that growth is.