NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

Zhipu's Hong Kong Debut Made It the First Major Chinese Generative-AI Startup to Go Public

Zhipu raised about HK$4.3 billion in a Hong Kong listing oversubscribed more than a thousand times, with 70% of proceeds earmarked for model research.

2026-11-17 · 1021 words · NeuroAI
Zhipu's Hong Kong Debut Made It the First Major Chinese Generative-AI Startup to Go Public

The bell rang at the Hong Kong Stock Exchange on a Thursday morning for a company that had spent seven years building models. The retail portion of its offering had been applied for more than a thousand times over. By the end of the session the shares sat well above the issue price — and the real test had only just begun.

On 8 January 2026, Beijing Zhipu Huazhang Technology Co., Ltd. (北京智谱华章科技股份有限公司), known internationally as Zhipu or Knowledge Atlas Technology, began trading under ticker 02513. Reuters described it as the first major Chinese generative-AI startup to go public. Whether that label is historic or merely convenient, the listing puts a company that sells foundation models in front of ordinary shareholders for the first time.

What was actually sold

Zhipu offered shares at HK$116.20 each. Gross proceeds came to more than HK$4.3 billion; the net figure cited by Chinese financial media was HK$4.173 billion (≈US$535 million). The company listed under Chapter 18C of the exchange's rules, the route created for specialist technology companies that have not yet reached conventional profitability milestones.

The investor list was heavy by design. Eleven cornerstone investors subscribed a combined HK$2.98 billion (≈US$382 million), including JSC International Investment Fund SPC, JinYi Capital Multi-Strategy Fund SPC, Perseverance Asset Management, Shanghai Gaoyi, Taikang Life and GF Fund. Those commitments covered close to 70% of the shares on offer before any greenshoe adjustment, which leaves a small float and a large block of holders with an interest in stability.

  • Offer price: HK$116.20 per share.
  • Retail demand: 1,159.46 times oversubscribed in the Hong Kong public offering.
  • Institutional demand: 15.28 times oversubscribed in the international tranche.
  • First trade: opened at HK$120, rose more than 13% to HK$131.5 (US$17) intraday.
  • Market value: about HK$57.89 billion (≈US$7.4 billion) at the Reuters-reported peak.

Read those numbers as sentiment, not as a verdict on the business. A heavily oversubscribed IPO with a tight float can move a long way on very little volume.

Where the money goes

The prospectus is unusually specific about the plan. According to the filing, 70% of the net proceeds will go to research and development of general-purpose AI large models (大模型) — pre-trained models, deep reasoning models, AI agents and the company's GLM framework. A further 10% is earmarked for improving the Model as a Service (MaaS) platform, 10% for building the partner network and making strategic investments, and 10% for working capital and general corporate use.

That allocation is a statement of strategy. Zhipu is not presenting itself as an application company chasing users. It is telling investors that the moat is the base model, and that the return will come when other companies build on it.

The company's roots help explain the pitch. Founded in 2019 by researchers from Tsinghua University, Zhipu released its GLM pretraining framework in 2021 and the GLM-130B open model in 2022. Zhang Peng serves as chief executive, Liu Debing as chairman, and co-founder Tang Jie as chief scientist. The prospectus describes a business selling to both institutions — private companies and public-sector bodies — and to individual developers.

The spending behind the listing

The reason for going public is visible in the research line. Citing the prospectus, Chinese financial press reported R&D spending of RMB 84.4 million in 2022, RMB 528.9 million in 2023 and RMB 2.1954 billion (≈US$306 million / HK$2.39 billion) in 2024. The first half of 2025 alone added RMB 1.5947 billion (≈US$222 million / HK$1.74 billion), bringing cumulative spending to roughly RMB 4.4 billion (≈US$613 million / HK$4.79 billion). The filing says about 74% of staff work in research and development.

That is the structural argument for the IPO. Training frontier models is a cash business with a long payback, and private venture funding has limits. A public listing gives a company with heavy fixed costs a recurring channel to capital — and gives the market the job of deciding, every quarter, whether the spending is justified.

It is also a risk. Public investors are less patient than strategic ones about losses, and a model roadmap measured in years does not fit neatly into quarterly reporting.

What a public listing changes

For the wider generative-media industry, the significance is not the share price. It is that a company whose product is a model is now subject to the same disclosure discipline as any other listed business. Revenue mix, customer concentration, compute commitments and research spend will be reported on a schedule, whether the numbers flatter the story or not.

That should be useful to anyone building on top of Chinese models. Developers have mostly had to judge foundation-model vendors through benchmarks, press releases and rumour. Filings are a different quality of evidence, even when they are carefully worded.

The debut also sets a reference point for the next wave of AI companies considering Hong Kong. If public markets reward the model-first story, more will follow. If they punish it, the financing conversation changes across the sector.

Honest limitations

This article draws on Reuters reporting carried by China Daily Hong Kong, on Chinese financial-media coverage of the listing, and on figures attributed to the company's prospectus. I have not read the full prospectus independently, and several numbers here are second-hand transcriptions of it. Intraday prices and market capitalisation move quickly; the figures describe the first session, not the current position. The "first" framing comes from Reuters and from the company's own promotional language, and depends on how one defines a generative-AI startup. I have not assessed the company's revenue quality, its compute supply, or the valuation against peers, and I make no judgement on whether the shares are worth buying.

What readers can do now

  • If you build on a foundation model, watch the listed company's disclosures. They will tell you more about its commitments and customer base than any benchmark score.
  • Read an IPO allocation paragraph as a strategy statement. A 70% research earmark says this company intends to compete on the base layer, not on packaging.
  • Treat first-day oversubscription as a measure of attention, not of product quality. Come back when the first full reporting cycle is published.

Related coverage

Cite this article

NeuroAI (脑机网), "Zhipu's Hong Kong Debut Made It the First Major Chinese Generative-AI Startup to Go Public," 17 November 2026, https://neuroai.site/a/na-aigc-zhipu-hong-kong-ipo

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Generative Media” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories