Zhipu AI and MiniMax posted broadly similar first-half results, but their business models are rapidly converging around API revenue.
By Wang Yuanlei
For two listed Chinese AI companies with nearly identical revenues, the market has assigned starkly different valuations — at one point separated by more than HK$400 billion ($51 billion). Z.AI (Zhipu AI) (2513.HK) reported its 2026 interim results on Aug. 31, following MiniMax’s (0100.HK) disclosure several days earlier.
In the first half, Zhipu generated 954 million yuan ($142 million) in revenue, while MiniMax reported $117 million, a gap of less than 15%. Both remain deep in the red — Zhipu posted a loss of 2.07 billion yuan while MiniMax recorded an adjusted net loss of 2.1 billion yuan.
The two companies began from nearly opposite positions just a year ago but have converged on the same strategy: application programming interface (API) sales, where developers pay for usage.
Zhipu: from project-based to API-first
Last year Zhipu resembled a project-based software firm. Its 2025 annual report showed localized deployment revenue jumped 102.3% to 534 million yuan last year, accounting for 73.7% of total revenue — essentially selling models as one-off deliveries.
By the first half of this year, local deployment had fallen to 129 million yuan, or 13.5% of revenue, at drop of 20.5% from a year earlier. Zhipu said that as cloud deployment has grown rapidly, it is placing greater emphasis on the quality of localized business and its choice of industries.
Meanwhile, “selling usage” has become the core business. Open platform and API revenue reached 825 million yuan, a 2,735.7 per cent year-on-year surge, jumping from 15.2% of total revenue to 86.5%. By contrast, enterprise general-purpose models, which contributed 148 million yuan, or 77% of revenue a year ago, shrank to 67 million yuan, or 7%.
Zhipu described this as a natural consequence of advances in model capabilities. The implication: with sufficiently powerful models, clients prefer pay-per-use pricing over expensive upfront purchases.
MiniMax: from consumer to enterprise
MiniMax initially focused on consumer-facing products. Though consumer AI revenue fell in the first half, AI-native products still brought in about 308 million yuan, or 36.6% of total revenue.
The bigger story is elsewhere. Open platform and other AI enterprise services generated around 534 million yuan of revenue, up from 66 million yuan a year earlier — a 703% increase — making it MiniMax’s largest revenue source. Like Zhipu, MiniMax is now enterprise-led.
The two financial reports point to a broader shift in the AI industry. Revenue is moving away from projects and subscriptions toward usage-based pricing, increasingly resembling utilities such as water, electricity and gas.
The model becomes infrastructure; API calls become the flow through that infrastructure. Customers pay for what they use.
This mirrors the industry’s evolution: from selling models to selling usage, then subscriptions, and eventually end-to-end task outcomes. As models become capable of completing more of a task independently, customers increasingly pay for the result rather than the underlying model. The unit used to measure revenue changes accordingly.
Rapid growth doesn’t equal easy money
Growth rates and the quality of growth deserve scrutiny. Zhipu’s 2,735.7% API surge came off a meagre base of just 29.1 million yuan. Notably, average API prices rose about 101%, suggesting revenue growth is not purely volume-driven at the expense of pricing.
The shift toward APIs is not unique to China. Overseas peers are following a similar path. More than 80% of Anthropic’s revenue, for example, comes from enterprise APIs and developers. China’s AI model companies are therefore moving toward a business model that has already gained traction internationally.
Despite comparable revenues, valuations diverge wildly. At the market close on Sept. 1, Zhipu’s market capitalisation stood at HK$554.1 billion versus MiniMax’s HK$121.3 billion — a gap of over HK$430 billion. At its peak, Zhipu’s market value briefly exceeded HK$1 trillion, reflecting what some call market irrationality.
Anthropic is reportedly targeting a late-September or early-October initial public offering, with potential valuations exceeding $2 trillion. For AI companies, there seems to be no upper limit to valuations. Perhaps the simplest explanation is that markets do not always follow conventional rules.
The acid test
Beyond revenue, two other figures in the companies’ financial reports deserve attention. The first is research and development spending.
Zhipu spent 2.13 billion yuan on R&D in the first half, up 33.6% from a year earlier. That was more than double its revenue for the period. MiniMax showed a similar pattern, spending about 2.14 billion yuan against revenue of $117 million. The AI model arms race is far from over.
The second issue is profitability.
Zhipu’s overall gross margin was 26.4% in the first half, compared with 17.9% for MiniMax. Compared with traditional software businesses, API-based AI looks more like a volume-driven business in which scale must compensate for relatively modest pricing power.
That’s hardly a comfortable business.
Taken together, the earnings reports show that China’s AI model companies have successfully established “selling usage” as a viable revenue model. The numbers make that transition unmistakable.
The next challenge will be converting those API calls into profits. That will be the real test.
Source:
EO Intelligence
