December 19, 2025, marks three years since ChatGPT's debut and a year after China's "Hundred-Model War" largely concluded. On this day, the Hong Kong Stock Exchange (HKEX) updated its website with a listing application from Zhipu AI, a Chinese large model company, signaling its entry into the capital markets. Could this mean the arrival of the world's first large model stock?
Zhipu's growth trajectory is steep. Revenue surged from RMB 57.4 million in 2022 to RMB 124.5 million in 2023 and RMB 312.4 million in 2024, achieving a 130% CAGR. Frost & Sullivan reports Zhipu as China's largest independent large model provider by 2024 revenue.
Currently, Zhipu's business model leans "traditional," primarily serving enterprise (B2B) and government (G2B) clients via two revenue streams: on-premise deployment (private cloud) and cloud-based MaaS (Model-as-a-Service) token calls. Initially, 95.5% of revenue came from private deployments, but this share dropped to 84.8% by mid-2025, with MaaS contributing 15.2%. CEO Zhang Peng aims for a 50-50 split.
Despite shifting revenue streams, gross margins remained stable above 50%, reflecting efficient model delivery without heavy infrastructure burdens. However, R&D spending defines Zhipu's core. Expenditure skyrocketed from RMB 84.4 million (2022) to RMB 5.289 billion (2023), RMB 21.954 billion (2024), and RMB 15.947 billion in H1 2025. With 657 R&D staff (74% of total), Zhipu expanded beyond language models to multimodal, agent, and edge models, exemplified by its GLM-4.6 update.
Adjusted net losses widened from RMB 97 million (2022) to RMB 6.21 billion (2023), RMB 24.66 billion (2024), and RMB 17.52 billion in H1 2025, driven by R&D. Sales and administrative costs, however, narrowed proportionally.
Originally targeting China's A-share market, Zhipu pivoted to an A+H dual listing. If successful, it would debut an unproven business model for capital markets to price—a mix of private deployments, cloud MaaS scaling, and relentless R&D.
Zhipu’s journey began in 2022 amid pre-boom skepticism. Early funding rounds (Angel in January, Series A in May) valued it at RMB 152 million for a RMB 2.28 million registered capital tranche, signaling rapid valuation spikes. Despite heavy reliance on G2B/B2B deals, Zhipu pursued open-source influence, releasing GLM-130B and CodeGeeX. Post-ChatGPT, it launched ChatGLM, joining the conversational AI race.
By 2024, industry divergence saw Zhipu pivot to multimodal products (voice, vision, video). Yet 2025’s shakeout, intensified by DeepSeek’s open-source aggression, pushed peers toward applications. Zhipu doubled down on foundational models, aligning with its AGI roadmap—China’s first 10B/100B models, open-source trillion-parameter model, and device-control agent.
Pre-IPO, Zhipu streamlined focus: scaling back multimodal efforts to prioritize text/code models while boosting API/cloud services. Top-five client revenue share fell from 45% (2024) to 40% (H1 2025), reducing dependency.
**Betting on APIs** Though 85% reliant on private deployments in 2024, Zhipu is pivoting to MaaS. CEO Zhang revealed GLM Coding Plan’s global ARR surpassed RMB 100 million (USD 14 million), highlighting API growth. Post-GLM-4.5, MaaS revenue surged, driven by code model calls. On OpenRouter, GLM-4.5/4.6 outperformed all domestic models in paid traffic. Southeast Asia’s sovereign model deals also contributed.
Despite cloud giants’ MaaS homogenization, Zhipu secured nine of China’s top ten internet firms (e.g., Tencent, ByteDance) for code tools, proving demand for third-party models. Daily token consumption exploded from 500 million (2022) to 4.6 trillion (H1 2025), a key MaaS metric.
**Industry Awaiting Valuation** The sector lacks a consensus business model. From "Hundred-Model War" to today’s few pre-training stalwarts, paths remain divergent. C2C models face high compute costs; B2B-heavy models sacrifice scalability. Zhipu’s hybrid approach—private deployments for stability, MaaS for growth—offers a middle ground, though unproven at scale.
As the potential first listed large model firm, Zhipu’s valuation, investor focus (e.g., API scalability vs. cash flow), and market response will set industry benchmarks. Its IPO represents not just a corporate milestone but a litmus test for pricing an evolving sector.
With base model capabilities rising, API calls are exploding—OpenRouter’s annual tokens grew 10x (10T to 100T) from 2024 to mid-2025, signaling AI’s transition from experimentation to adoption. Zhipu’s listing may redefine how capital markets value this nascent yet transformative industry.
Comments