Zhipu Grows Revenue and Users on Path to China’s First AI IPO

Zhipu, a six-year-old Chinese AI startup, is generating over 100 million yuan (~$14m) in annual recurring revenue from selling access to developer tools and models, CEO/co-founder Zhang Peng said. Usage of its subscription service rose tenfold in the two months after the July launch of flagship model GLM-4.5, and the company is positioning the fast-growing tools business to support an upcoming IPO as it seeks to convert Chinese customers to paid AI services.

Analysis

Market structure: Zhipu’s ability to reach >100M CNY ARR and 10x usage after GLM-4.5 shows willingness of Chinese developers to pay for AI tools, favoring cloud/AI-infrastructure providers (cloud, inference chipmakers, model-hosting platforms) while pressuring low-margin consumer app monetization. Expect incremental pricing power for domestic cloud players (Alibaba BABA, Baidu BIDU, Kingsoft Cloud KC) but continued cost pressure on GPU supply (benefitting NVDA) and potential upward pressure on high-end semiconductor demand over 6–18 months.

Risk assessment: Tail risks include a China-specific regulatory clampdown on model licensing or data rules and renewed US export controls on inference GPUs; both could halve TAM in a stress scenario within 3–12 months. Near-term (days–weeks) volatility will track IPO timelines and model adoption metrics; medium/long term (6–24 months) risks are monetization (ARPU <¥1k/year) and incumbent retaliation; hidden dependencies include domestic chip/cloud supplier capacity and government procurement preferences.

Trade implications: Prefer concentrated, conviction-weighted exposure to AI infra: small longs in BIDU and BABA (cloud/AI units) and NVDA for hardware; use 6–12 month call spreads to cap cost. Implement relative-value: long BABA cloud (+2–3% portfolio) vs short JD (JD) (-1–1.5%) to express AI infra vs retail spend divergence. Time entries around Chinese earnings/IPOs (act within 30–90 days), trim winners at +30% and cut losers at -20%.

Contrarian angle: The market may over-index on headline ARR growth and underweight CAC/churn; past China AI winners (SenseTime, iFlytek) show high early ARR but weak public-market profitability for 12–36 months. If Zhipu’s IPO is priced aggressively, short-window arbitrage (buy put spreads or sell covered calls) could profit from mean reversion; conversely, underappreciated enterprise spend could surprise positively if government adoption accelerates.

Key Decisions for Investors

• Establish a 2–3% long position in Baidu (BIDU) to capture China search-to-cloud AI monetization; use a 9–12 month calendar call spread (e.g., buy 12-month ATM call, sell 6-month higher-strike call) to limit premium; take profits at +30% or exit on a -20% drop.
• Allocate 1–2% to NVIDIA (NVDA) via a 6–9 month bull-call spread (buy 1–2% delta OTM call, sell 10–15% higher strike) to express GPU demand for inference; target a 6–18 month horizon and reduce if US export controls tighten within 30–90 days.
• Initiate a pair trade: long Alibaba (BABA) cloud exposure 2% vs short JD.com (JD) 1–1.5% to express AI infra upside vs retail cyclicality; rebalance after Chinese cloud earnings or Zhipu IPO filing (monitor next 30–60 days), trim net exposure if the spread narrows <5% relative performance.
• If Zhipu announces IPO pricing or guidance, deploy a short-term volatility play: buy 1–2% notional 3–6 month put spreads on the IPO (or related small-cap China AI ETF) to hedge downside from an over-priced debut; unwind on IPO day or if post-IPO lockup creates >25% move.
• Reduce exposure to large-cap consumer internet names lacking cloud/AI moats by 1–2% (e.g., Meituan 3690.HK, PDD) and redeploy into AI infrastructure over 3–12 months; reassess if regulatory signals relax within 90 days.