💰钛媒体•Stalecollected in 2h
Zhipu AI-backed startup eyes record-breaking IPO
💡See how Zhipu AI's investment strategy is scaling industrial AI startups to public markets.
⚡ 30-Second TL;DR
What Changed
Zhipu AI achieves its first portfolio company IPO
Why It Matters
This IPO signals strong market confidence in AI-related infrastructure and service providers backed by major Chinese LLM labs.
What To Do Next
Monitor Zhipu AI's investment portfolio to identify emerging AI-integrated industrial service providers.
Who should care:Founders & Product Leaders
Key Points
- •Zhipu AI achieves its first portfolio company IPO
- •The startup is on track to set a record for the fastest IPO
- •Highlights the growing influence of Zhipu's venture capital arm
🧠 Deep Insight
Web-grounded analysis with 18 cited sources.
🔑 Enhanced Key Takeaways
- •Zhipu AI, officially known as Knowledge Atlas Technology and branded internationally as Z.ai, became the first large language model (LLM) company globally to go public, listing on the Hong Kong Stock Exchange on January 8, 2026.
- •The company successfully raised approximately $558 million to $560 million USD in its initial public offering (IPO), achieving an initial market valuation estimated between $4.4 billion and $7.5 billion USD.
- •Zhipu AI's core business model is centered on a Model-as-a-Service (MaaS) platform, primarily targeting enterprise deployments, and it offers a range of open-source models within its GLM series, some of which are notably trained entirely on Huawei Ascend chips.
- •Despite reporting significant revenue growth, reaching RMB 724 million ($104.8 million) in 2025, Zhipu AI also incurred substantial net losses of RMB 4.72 billion in the same year, reflecting a strategic prioritization of Artificial General Intelligence (AGI) advancement and market share over immediate profitability.
- •In early 2025, Zhipu AI was placed on a U.S. government trade blacklist, a development that could pose challenges to its global expansion efforts and access to advanced hardware crucial for model training.
🛠️ Technical Deep Dive
- GLM-5: Utilizes a Mixture-of-Experts (MoE) architecture with a total of 745 billion parameters, of which 44 billion are active per inference operation. It features 256 experts, with 8 activated per token. The model supports context windows up to 200,000 tokens and incorporates the DeepSeek Sparse Attention mechanism for efficient long-sequence processing. Notably, GLM-5 was trained exclusively on Huawei Ascend chips and supports inference frameworks like vLLM and SGLang.
- GLM-4.5 Series: Includes GLM-4.5 (355 billion parameters) and GLM-4.5-Air (106 billion parameters), both leveraging a Mixture of Experts architecture. These models support hybrid reasoning modes ('thinking mode' for complex tasks and 'non-thinking mode' for speed) and feature Multi-Token Prediction for rapid inference, achieving up to 200 tokens/sec.
- GLM-4.6: Marked the first integration of FP8 and Int4 quantization on Cambricon chips and offers native FP8 support on Moore Threads GPUs.
- Ying: A text-to-video model capable of generating six-second video clips from text and image prompts in approximately 30 seconds.
- GLM-130B: A 130-billion-parameter bilingual (Chinese-English) large language model.
- GLM-4.1V-Thinking: A next-generation general visual language model designed with enhanced reasoning capabilities.
- Open-Source Strategy: Many of Zhipu AI's models, including GLM-5, GLM-4.7, GLM-4.5, GLM-130B, and ChatGLM, are released under the permissive MIT License, allowing for unrestricted commercial use, adaptation, and further development.
🔮 Future ImplicationsAI analysis grounded in cited sources
Zhipu AI's global expansion and 'sovereign AI' initiatives will intensify geopolitical competition in the AI sector.
The company is actively building AI infrastructure for 10 ASEAN and 10 Belt and Road nations, directly challenging Western AI dominance and potentially creating new technological blocs.
The company's aggressive R&D spending and stated focus on 'super cognitive intelligence beyond human level' will continue to drive significant financial losses in the short to medium term.
Zhipu AI has explicitly stated that profitability is not a priority, with R&D spending far exceeding revenue, indicating a long-term investment strategy towards achieving AGI.
Zhipu AI's reliance on domestic Chinese chips will accelerate China's self-sufficiency in AI hardware despite U.S. trade blacklists.
The company has demonstrated successful training of its advanced models, such as GLM-5, entirely on non-NVIDIA hardware like Huawei Ascend chips, showcasing a viable alternative supply chain for AI development.
⏳ Timeline
2019
Zhipu AI founded as a spin-off from Tsinghua University.
2021
Co-founder Tang Jie helped develop the Wu Dao 2.0 model (1.75-trillion parameters) at the Beijing Academy of Artificial Intelligence (BAAI).
2022
Zhipu AI and Tsinghua's lab jointly debuted GLM-130B, a 130-billion-parameter bilingual model.
2025-01
Placed on a U.S. government trade blacklist.
2025-07
Released GLM-4.5 and GLM-4.5 Air, rebranding internationally as Z.ai.
2026-01-08
Officially listed on the Hong Kong Stock Exchange, becoming the world's first public company focused on general AI foundation models.
📎 Sources (18)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 钛媒体 ↗