科创板对AI大模型敞开大门：不盈利也能上市，资本市场开始为技术信仰定价 | China's STAR Market Opens to AI Model Companies: No Profit Required for IPO, Markets Begin Pricing Tech Conviction
6月17日陆家嘴论坛上，中国证监会主席吴清宣布了一个被很多人低估的消息：科创板第五套上市标准正式扩围至人工智能大模型领域。同日，上交所发布了《人工智能大模型企业适用科创板第五套上市标准审核指引》。
简单说就是：没盈利的AI大模型公司，现在可以在A股上市了。
“第五套标准”到底意味着什么？
第五套标准原本是为医药研发企业设计的——允许没有收入、没有利润的创新药企上市，因为新药研发周期长达10年，不可能在获批前盈利。这套逻辑被搬到了AI大模型领域。
《指引》里几个关键门槛：最近三年累计研发投入不低于3亿元；预计市值不低于40亿元；核心技术人员稳定；主要产品的技术壁垒有充分论证。没有盈利要求，没有营收下限。你在烧钱搞研发？没问题，只要证明你在做”真的技术创新”而非”PPT圈钱”。
这个变化有多大？2023年之前，一家AI公司想在国内IPO，要么借壳要么等到盈利，两条路都不好走。现在科创板直接给你开了扇门。
为什么是现在？
时间点耐人寻味。2026年上半年，中国AI行业出现了一个独特现象：DeepSeek、智谱、百川等一批大模型公司在技术上已经具备全球竞争力，但融资渠道仍然狭窄——要么美元基金（受地缘政治干扰），要么地方政府引导基金（规模有限）。
资本市场缺位的结果是：中国AI公司的估值逻辑长期被美国的OpenAI、Anthropic的融资新闻主导。你的技术指标追上了GPT-5.5，但你的估值只有对方的零头。这不是技术问题，是资本市场定价权的缺失。
科创板扩围是一次信号释放：中国资本市场要开始独立给AI公司定价了。
市场准备好了吗？
但问题也摆在那。A股散户为主的结构，能不能接住”不盈利+高研发+长周期”的AI公司？纳斯达克花了三十年才建立起对科技股的定价框架——亚马逊亏了快20年，股价却涨了上千倍，因为投资者理解了”亏损是为了增长”这个逻辑。
A股散户看到一家公司”收入为0但研发花了5个亿”，第一反应可能是”这公司有毛病吧”而非”这是技术壁垒在积累”。
这个认知鸿沟需要时间填平。科创板的机构投资者门槛已经很高（50万起），这在一定程度上有隔离效果。但AI大模型公司一旦挂上”高科技+不赚钱”的标签，股价波动幅度会远超消费股，对监管的考验也在后面。
上交所显然意识到了这个问题。《指引》特别强调”技术壁垒的充分论证”——需要第三方技术评估报告、同行评议等材料。这是在用制度替代散户判断：让专家替市场把关”这家公司是真有技术还是编故事”。
信号意义大于融资意义？
还有一个容易被忽略的细节：第一批符合条件的AI大模型公司可能不超过10家。DeepSeek、智谱、MiniMax、月之暗面、百川智能、阶跃星辰、零一万物——两只手数得过来。
所以短期内，科创板大模型IPO不会是一场”上市潮”，而是一批精选公司的”精品首发”。真正的意义在于定价体系的建立——当第一家AI大模型公司在A股挂牌，它的股价走势、市盈率（市盈率=股价/每股收益——等等，它没有收益，所以是市研率）、研发投入的市场反馈，会成为整个赛道的估值锚。
这对早期风投退出、对AI创业者的创业方向选择、甚至对大学生选专业，都会产生传导效应。
资本市场开始为技术信仰定价——这个变化值得关注。
China’s STAR Market Opens to AI Model Companies: No Profit Required for IPO
On June 17, at the Lujiazui Forum, CSRC Chairman Wu Qing announced a move many underappreciated: the STAR Market’s fifth listing standard now covers AI large model companies. The Shanghai Stock Exchange simultaneously published detailed review guidelines.
Translation: unprofitable AI model companies can now IPO on China’s A-share market.
What Does the “Fifth Standard” Mean?
The fifth standard was originally designed for pharmaceutical R&D companies — allowing innovative drug makers with zero revenue and zero profit to go public, since drug development takes 10+ years. That logic is now applied to AI.
Key thresholds in the guidelines: cumulative R&D spending of at least ¥300 million over three years; expected market cap no less than ¥4 billion; core technical team stability; rigorous proof of technical moat. No profit requirement. No revenue floor. Burning cash on R&D? Fine, as long as you demonstrate real technological innovation, not PowerPoint fundraising.
This is significant. Before 2023, an AI company wanting to list domestically either had to backdoor-list or wait for profitability — both painful paths. Now there’s a direct door.
Why Now?
The timing is telling. In H1 2026, China’s AI sector hit a paradox: companies like DeepSeek, Zhipu, and Baichuan are globally competitive on technical benchmarks, but their funding channels remain narrow — either USD funds (geopolitically constrained) or local government-guided funds (limited scale).
Absent capital market access, Chinese AI company valuations have been dictated by OpenAI and Anthropic’s fundraising headlines. Your technical metrics match GPT-5.5, but your valuation is a fraction of theirs. That’s not a tech problem — it’s a missing capital market pricing mechanism.
The STAR Market expansion is a signal: China’s capital markets are starting to price AI companies independently.
Is the Market Ready?
The challenge is obvious. A market dominated by retail investors — can it handle “unprofitable + high R&D + long cycle” AI companies? Nasdaq took 30 years to build its tech stock pricing framework. Amazon lost money for nearly 20 years while its stock rose 1000x, because investors understood “losses fund growth.”
A Chinese retail investor seeing “revenue: ¥0, R&D: ¥500M” might think “this company is broken” rather than “this is a technology moat being built.”
This cognitive gap will take time. STAR Market’s high institutional investor threshold (¥500K minimum) provides some buffer. But AI model companies tagged “high-tech + no profit” will see far more volatility than consumer stocks — a regulatory stress test lies ahead.
The SSE clearly anticipated this. The guidelines emphasize “rigorous proof of technical moat” — requiring third-party technical assessment reports, peer reviews, and more. This replaces retail judgment with institutional gatekeeping: let experts verify whether a company has real technology or just a story.
Signaling Matters More Than Fundraising
One easily overlooked detail: the first batch of eligible AI model companies likely numbers fewer than 10. DeepSeek, Zhipu, MiniMax, Moonshot AI, Baichuan, StepFun, 01.AI — you can count them on two hands.
So the near term won’t bring an “IPO wave” but a curated set of “premium debuts.” The real significance is establishing a pricing system. When the first AI model company lists on the A-share market, its stock trajectory, R&D multiple, and market feedback on R&D spending will become the valuation anchor for the entire sector.
This ripples outward — to VC exits, to entrepreneurial choices, even to what majors college students pick.
Capital markets beginning to price technological conviction — that’s a change worth watching.