Wire flash
Zhipu AI market cap plunges to 300 billion HKD amid LLM price war
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
The article reports on the escalating price war in the large language model (LLM) sector, focusing on Chinese AI company Zhipu AI (智谱). Zhipu's stock price has fallen sharply from a June high of 2,980 HKD to around 610 HKD, with its market cap dropping to approximately 300 billion HKD. The decline coincides with aggressive price cuts by both domestic and international AI model providers. In September 2024, companies including Zhipu, Xiaomi, and OpenAI announced significant reductions in API pricing for their models. OpenAI's GPT-6 Sol and Luna API prices are 50% lower than previous models. Despite Zhipu's strong revenue growth (399.7% year-on-year to 954 million yuan in H1 2024) and a 2,735.7% surge in API business revenue, the company reported a net loss of 2.071 billion yuan due to high R&D spending and low margins on cloud deployment. Analysts cited in the article predict that low-end API prices will continue to fall to $0.1-0.2 per million tokens by H2 2026, with some Chinese firms potentially subsidizing prices at zero or negative margins. Goldman Sachs is quoted as saying lower model costs will further drive AI adoption.
Source report
A price war in the large language model (LLM) sector is escalating, as both domestic and international players slash API costs.
Zhipu Stock Volatility
On September 25, Zhipu's Hong Kong-listed stock continued its downward trend, dipping to around HK$610.5 during intraday trading. By market close, the decline narrowed to 0.78%, bringing the company's total market capitalization back to approximately HK$300 billion.
Zhipu's stock has experienced extreme volatility this year:
- January 8: Listed at an issue price of HK$116.2
- June 22: Surged to an all-time high of HK$2,980 per share
- Mid-July: Fell below HK$1,000, later rebounded above HK$1,200
- Recent: Dropped back to around HK$600
Global Price Declines in Model Inference
Alongside the stock correction, global LLM inference prices have fallen sharply.
Domestic Players
- September 10: Zhipu released GLM-5.3-Flash and adjusted API prices. Peak-hour pricing is now RMB 2 per million tokens for cache-missed input and RMB 8 for output, with off-peak rates halved.
- September 22: Xiaomi launched the MiLM series. The flagship Pro version charges RMB 3 and RMB 6 per million tokens for uncached input and output, respectively. The Flash version is further reduced to RMB 1 and RMB 2. With cache hits, Pro input costs only RMB 0.025 per million tokens.
International Players
- September 22: OpenAI released GPT-6 Sol and Luna, claiming a 40% cost reduction on typical tasks compared to GPT-5 Opus. API input and output prices are $4 and $20 per million tokens (approximately RMB 26.9 and RMB 134.3), with cache read prices down 60%.
- Same day: OpenAI also launched GPT-6 Sol and Luna. Luna's standard API input and output prices are $0.1 and $0.5 per million tokens (approximately RMB 0.67 and RMB 3.36). OpenAI stated that by improving cache and inference efficiency, GPT-6 Sol and Luna API prices are 50% lower than the discounted rates for GPT-5.6 equivalent models.
Market Impact on Zhipu
Since DeepSeek announced a new round of price adjustments for its Flash series models in September, Zhipu's stock has faced significant pressure, gradually declining to around HK$600. This reflects market concerns over intensifying competition.
Currently, Zhipu's pricing is as follows:
- GLM-5.3-Flash: RMB 0.8 per million tokens for uncached input, RMB 2.8 for output, and RMB 0.23 for cached input
- GLM-5.3 (Flagship): RMB 8 and RMB 28 per million tokens for uncached input and output, respectively
Financial Performance
Zhipu previously improved profitability by increasing average selling prices alongside rapid growth in call volumes. The company's semi-annual report for the first half of the year showed:
- Revenue: RMB 954 million, up 399.7% year-on-year
- Open platform and API revenue: RMB 825 million, up 2,735.7% year-on-year
- MaaS platform token calls: Increased more than 40 times compared to the beginning of the year
- Average API selling price: Increased by approximately 101%
- Gross margin for open platform and API business: Rose from -0.4% to 24.6% year-on-year
However, overall gross margin fell from 50% to 26.4% year-on-year. R&D spending reached RMB 2.131 billion, with a net loss attributable to shareholders of RMB 2.071 billion. Zhipu attributed this primarily to the rapid expansion of its cloud deployment business, which is still in the gross margin ramp-up phase.
Computing Power Constraints
Insufficient computing power has been a bottleneck for Zhipu's commercial expansion.
A Zhipu executive revealed on September 16 that after the GLM-5 launch in February, model call demand quickly grew to about 10 times previous levels, rapidly depleting computing reserves. The company had to temporarily suspend sales of its Coding Plan. After completing approximately $4 billion in refinancing in July, Zhipu began increasing computing investments, and the Coding Plan was fully reopened for sale. The executive noted that post-revival sales grew more than 15 times.
New Funding Round
As model call demand continues to grow, Zhipu launched a new funding round in September, issuing new shares at HK$714 per share along with zero-coupon convertible bonds, raising approximately HK$39.3 billion (about RMB 33.67 billion). The new share placement price represented a discount of approximately 9.96% from the closing price before the announcement. Combined with the approximately HK$31.4 billion (about RMB 26.89 billion) raised in July, Zhipu's total funding for the year has exceeded HK$70 billion.
Industry Outlook
Analysts suggest that domestic models like Zhipu previously gained market share through cost-effectiveness in a competitive landscape. However, as major domestic and international model providers join the price-cutting trend, "low price" is no longer a unique competitive advantage but is becoming an industry norm.
Recent research indicates that models with cutting-edge performance and multimodal capabilities still maintain strong pricing power, but the low-end API market has entered a price war. It projects that by the second half of 2026, low-end API prices will range from $0.1 to $0.2 per million tokens (approximately RMB 0.67 to RMB 1.34), with continued pressure on prices and gross margins. Some well-funded Chinese model makers may subsidize prices at zero or even negative gross margins in the short term. Meanwhile, Goldman Sachs believes that lower model costs will further drive AI adoption rates.
Source
腾讯网-股票Eastern
Part of this Story
Zhipu AI raises 756 billion HKD in 2026, market cap plunges 70% amid value reassessment