Chinese artificial intelligence companies Zhipu and MiniMax, both listed in Hong Kong, are entering a phase of rapid growth driven by low-priced API services, but investors are increasingly skeptical about the sustainability of their valuations. According to TMTPost, the two companies have seen their share prices fall sharply in September, even as they report triple-digit revenue growth.
On September 10, DeepSeek released its V4.1 Flash model and cut API prices again, with off-peak cache-hit input pricing dropping to 0.02 yuan per million tokens, uncached input to 1 yuan, and output to 4 yuan. This move underscores a broader industry trend: model capability improvements are now being used to justify price cuts rather than premium pricing. The same day, Zhipu's shares fell 10.34% and MiniMax fell 8.98%. By September 11, both stocks had lost over 20% since the start of the month.
Zhipu reported first-half 2026 revenue of 954 million yuan (about $134 million), up 399.7% year-on-year, surpassing its full-year 2025 revenue. MiniMax reported revenue of $116.6 million, up 283.1%, also exceeding last year's total. However, both remain loss-making, with Zhipu posting a net loss of 2.072 billion yuan (about $292 million) and MiniMax an adjusted net loss of $293 million.
Revenue shift to API services
The two companies have pivoted sharply toward open-platform and API services. Zhipu's open platform and API business generated 825 million yuan (about $116 million) in the first half, up 2,735.7% year-on-year, now accounting for 86.5% of total revenue. MiniMax's open platform and other AI enterprise services brought in $73.9 million, up 703.1%, representing 63.4% of revenue. This shift from one-time deployments and AI-native products to token-based consumption has accelerated growth but exposed revenue to price fluctuations.
At earnings calls, Zhipu disclosed an August annualized recurring revenue (ARR) of about $1.6 billion, while MiniMax reported over $800 million. ARR, which annualizes the latest month's revenue, has become a key metric, but it can swing sharply with price, customer mix, and call volumes. Jefferies, a Wall Street investment bank, raised its 2026-2029 revenue forecasts for Zhipu by 37% to 119% while cutting its loss estimates, yet lowered its target price from HK$1,299.80 to HK$1,183.79. The bank also cut the valuation multiple for Zhipu's open platform and API business from 50 times expected 2026 ARR to 30 times.
The valuation multiple cut reflects a market reassessment. Earlier, investors were willing to pay high multiples for rapid growth, but now they question whether revenue can sustain its price. A reduction from 50x to 30x ARR implies a 40% devaluation, requiring ARR to grow about 67% to compensate. Zhipu does retain some pricing power—it raised API prices by 83% in Q1 2026 while call volumes grew 400%—but the launch of cheaper Flash models, including its own GLM-5.3-Flash, is pressuring prices. The company reported that token call volumes on its MaaS platform grew over 40 times from the start of the year to August, with average API prices up 101% and paying daily active users up 603%.
Competition intensifies
Competitors are also entering the low-price segment. Alibaba launched Qwen3.8-Flash on the same day as Zhipu's GLM-5.3-Flash, followed by Tencent and DeepSeek with lightweight, high-throughput models. These models target high-frequency tasks like code development and office agents, where speed and cost matter more than top-tier capability. The price gap between models is narrowing, especially for cache input, which is 50 times cheaper than first-time processing. This means that if much of the call volume comes from low-priced Flash models and cached content, API revenue may not grow at the same pace.
The shift has broader implications for independent model makers. Unlike tech giants with cloud platforms and distribution channels, Zhipu and MiniMax must prove they can retain customers, expand call volumes, and improve margins amid falling prices. Their high growth is not over, but the capital market is no longer paying solely for growth. As TMTPost notes, the "low-price growth" era has arrived, and valuation multiples are being reset accordingly.
