Most AI founders sell a model. Dai Wenyuan (戴文渊) sold a different idea: that the valuable part of AI is not the chatbot, but the decision — the call a bank, a hospital, or a factory makes a thousand times a day.
That bet made 4Paradigm (第四范式) China's largest platform for "decision AI," and it is now being tested by the biggest shift in the field: the rise of the large model (大模型).
Building the decision-AI platform
Dai founded 4Paradigm in Beijing in September 2014, taking the roles of chairman, chief executive, and general manager. The company's pitch was "platform-centric" AI: give enterprises a system to build their own decision-making applications, find patterns in data, and improve operational choices — across finance, retail, manufacturing, energy, telecom, transport, and healthcare.
The strategy worked on paper. By 2022 revenue, 4Paradigm held about 22.6% of China's platform-centric decision-AI market, well ahead of the next player, according to CICC prospectus-cited research. It also repeatedly led China's machine-learning software market share in IDC rankings.
The Hong Kong listing
On 28 September 2023, 4Paradigm listed on the Main Board of the Hong Kong Stock Exchange under ticker 6682.HK. The offer price was HK$55.60 per share, with net proceeds of about HK$836 million. At opening the stock traded around HK$60, giving a market value above HK$28 billion, making it one of the larger TMT listings in Hong Kong that year.
Founder Dai directly held 22.88% of shares post-IPO, remaining the largest shareholder. Early backer Sinovation Ventures (创新工场), led by Kai-Fu Lee, had led the company's Series A in 2015 and followed on in later rounds — a rare case of a top Chinese VC staying through to a public listing.
From "Sage" to AIGS
Before the large model (大模型) wave, 4Paradigm's flagship was its "Sage" (式说) series; version 3.0 was positioned as a generative large model for enterprises. Dai's next move was a strategy he called AIGS — AI-Generated Software — using generative AI to rebuild enterprise software from the interface inward, so workers could instruct systems in natural language instead of clicking through rigid menus.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, 4Paradigm's survival hinges less on model quality than on whether enterprises trust an AI to actually make or shape high-stakes decisions.
The 2025 rebuild into a group
In 2025, 4Paradigm restructured into a group (later styled Phancy / 范式集团), with business lines spanning enterprise services, large model (大模型) and AGI (the "Phanthy / 万神殿" unit), consumer electronics, smart energy, and smart sports. The stated technical route shifted to "AI Agent + world model" — positioning an agent that acts in the real world rather than only answering questions.
The company reports more than 10,000 AI applications deployed across finance, retail, and healthcare, and points to a long client roster in regulated industries where data and compliance barriers protect incumbents.
Why this is a hard pivot
Decision AI was always the less flashy, higher-moat corner of the market — visuals and voice grabbed headlines, but banks and insurers pay for reliable calls. The large model (大模型) turn threatens that moat: general models can now do parts of what custom platforms once did. 4Paradigm's answer is to wrap its domain data and workflows around an agent layer, hoping its enterprise relationships and industry knowledge stay sticky even as raw model capability commoditizes.
The financials show the cost of the climb. Group revenue rose from RMB 942 million in 2020 to RMB 2.018 billion in 2021 and RMB 3.083 billion (≈ US$429 million / HK$24 billion) in 2022 — strong growth, but the company remained loss-making as it funded R&D and sales.
Why decision AI is the quiet corner that matters
Decision AI is easy to overlook next to chatbots and image models, but it is where enterprises actually spend: a bank's credit call or a factory's yield optimization repeats millions of times and touches real money. The large model (大模型) wave both threatens and helps 4Paradigm. It threatens because a general model can now draft parts of what custom platforms did; it helps because natural-language interfaces lower the adoption wall for non-technical staff. Dai's AIGS bet is to meet that duality head-on — wrap domain data and workflows around an agent layer so the platform stays sticky even as raw model capability commoditizes. The open question is whether enterprises will trust an agent to shape high-stakes calls, or keep it as a suggestion box.
Honest limitations
This profile is about enterprise AI and software, not brain-computer interface (脑机接口); neurotech readers should look to dedicated labs. 4Paradigm's post-2023 revenue, margin, and "10,000 applications" figures come from company disclosures and secondary summaries and were not independently audited here. The group restructuring and "AI Agent + world model" positioning are recent and unproven at scale; whether AIGS converts into durable enterprise revenue is still an open question. Dai's earlier academic and research background is lightly detailed in English-language sources and is kept brief rather than speculated upon.
What readers can do now
- Watch enterprise renewal rates, not launch slogans. The honest test for 4Paradigm is whether AIGS and agent products expand wallet share inside existing regulated-industry clients.
- Track the margin trend. Growth from RMB 3.08 billion (≈ US$429 million / HK$24 billion) in 2022 must convert to profit for the public-market story to hold.
- Compare the enterprise-AI lane. Against visual and embodied peers, 4Paradigm lives or dies on trust in AI-assisted decisions — the least visible, most consequential part of the large model (大模型) wave.