STAR Market's Fifth Listing Criteria Expanded to Large Model Companies, Opening IPO Path for DeepSeek and Others
N.R. Finch
China's securities regulator expanded the STAR Market's fifth listing standard to large-language-model companies, opening a direct A-share IPO path for still-unprofitable AI foundation-model developers like DeepSeek, Zhipu and MiniMax — a capital-market reform aimed squarely at the funding race behind global AI competition.
What is the fifth standard, and why extend it to AI now?
The fifth standard launched in 2019 for biotech firms — companies with long R&D cycles, heavy upfront spending, and no near-term profits that couldn't clear traditional earnings thresholds.
This means → LLM developers share almost the same financial profile: massive capital burn before commercialization, with revenue far short of R&D costs.
In plain terms = the rule was a door cut for "drug makers." Now "model makers" can walk through it too.
Not everyone qualifies — where is the bar?
The Shanghai Stock Exchange's guidance requires applicants to have at least one LLM product in scaled commercial deployment, plus demonstrated in-house R&D capability and a technology edge.
Regulators stated explicitly: the policy targets companies with real AI capability, not firms riding an AI narrative.
This means → "scaled commercial deployment" is a hard gate — a technical paper or a demo product is not enough. Real customers using it and real revenue flowing are required.
Which three companies benefit first?
DeepSeek: just closed its first external round at over ¥50 billion, with a valuation above ¥400 billion — a record for a Chinese LLM company. Founder Liang Wenfeng's equity structure is widely read as IPO groundwork.
Zhipu (智谱): plans to raise ¥15 billion for foundation-model R&D and an AI-services platform; has begun preparing a dual A-share / H-share listing.
MiniMax: has completed IPO coaching registration with Shanghai's securities bureau and formally entered the STAR Market listing process; revenue is growing at triple digits, but R&D spending still far exceeds revenue.
How much cash does the global LLM race burn?
Researchers estimate Google, Amazon, Meta, Microsoft and Oracle will spend a combined $670 billion+ on AI capex in 2026.
This reflects the nature of the LLM contest: spend first on compute, monetize later — whoever's funding chain breaks first exits first.
In plain terms = this reform plugs China's domestic AI firms into the public-market funding pipeline, so they are not left surviving on venture capital alone.
What is the real test ahead?
Industry observers say the reform aims to build a complete chain: venture capital → public markets → commercialization.
Whether the expanded fifth standard becomes a viable IPO path depends on each company meeting the core requirement of "scaled commercial deployment."
This means → the policy lane is open, but actually reaching the market still requires a product people pay for — the rule offers opportunity, not a guarantee.
Content is for reference only, not financial advice.