Ujian besar taruhan mendekat, perubahan besar pada pemimpin pintar bernilai 20 miliar

Reporter丨He Jipai Editor丨Chen Xiaoping

Boss of the academic superstar Dai Wenyuan, finally sees profit.

At the end of March, Paradigm Intelligence disclosed its performance, with revenue of 7.135 billion yuan in 2025, up 35.6%, and an adjusted net profit of 17.84 million yuan, turning losses into gains for the first time.

However, the overall capital market remains calm. As of April 2, market value is around 20 billion Hong Kong dollars, about one-third of its peak.

Paradigm Intelligence, formerly known as “Fourth Paradigm,” was founded by Dai Wenyuan, a top student from Shanghai Jiao Tong University, originally focused on vertical enterprise software models.

The booming large models have impacted Dai’s original architecture.

In the past year, he led the company’s rebranding and radical reforms, rebuilding the AI Platform, API, and Agentic AI as the “three pillars”, attempting to integrate Chinese chips, models, computing power, and even energy, to drive a new round of growth.

Dai Wenyuan’s new narrative for AI 2.0 is grand and filled with uncertainty.

Targeting the gaps

Paradigm Intelligence rebranded, and Dai Wenyuan specifically re-listed the financial statements.

The newly established three main businesses all have high AI concentration—focusing on PaaS with AI Platform, API business based on Token logic, and pay-per-result Agentic AI, each with its own focus and mutual empowerment.

Among them, AI platform business contributes over 90% of total revenue, which Dai regards as the ballast and power source.

Simply put, based on the traditional AI platform, it integrates three new products (ModelHub XC, HAMi vGPU, PhanthyCloud), achieving end-to-end, full-stack matrix delivery, mostly in the form of software-hardware integrated machines.

Dai Wenyuan’s focus is on one thing: bridging the gap between computing power, models, and applications.

The deployed models and underlying chip architectures are incompatible, and the hardware purchased lacks a sufficiently adapted model ecosystem. Different models often require separate adaptation, which is time-consuming and involves repetitive work, becoming a bottleneck for AI application explosion.

In this regard, an industry practitioner working on large models told 《21CBR》 that there are many domestic computing power brands, but maintaining model compatibility and ensuring domestically produced adaptation is a tough job.

“Currently, there is no particularly good solution,” he lamented, emphasizing the need for policy guidance to ensure mutual compatibility; otherwise, device costs are too high for software companies.

Where there is pain, there is demand.

Dai Wenyuan’s idea is to build an intelligent base, bringing together model manufacturers like Kimi and Qianwen, chip makers like Cambrian, and core clients such as central and state-owned enterprises and financial institutions, to align their needs.

On the left, based on ModelHub XC, creating the largest domestic trust innovation model community to improve model adaptation speed; on the right, using HAMi vGPU, enabling fine-grained scheduling of GPU computing power, like a CPU.

Above both, supported by PhanthyCloud for cloud support, and linked with Token services, allowing clients to use tokens to call cloud, models, and computing power on demand.

Paradigm has transformed into a “connector” for domestic computing power.

Dai Wenyuan’s new computing power business performs very well, contributing 6.552 billion yuan in revenue in 2025, up 32%. The unexpectedly rapid growth has given him great confidence.

When ModelHub XC was first released, the community model count was zero; by mid to late January, it exceeded 10,000, with an initial plan to reach 100,000 within a year, which has been achieved ahead of schedule.

Dai Wenyuan’s ambitions are high, aiming to catch up with Hugging Face by 2027, currently the world’s largest open-source AI community.

Paid incremental growth

After establishing the AI Platform infrastructure, Dai Wenyuan is also exploring more incremental businesses.

First, he plans to build a traffic pool around the API business to expand the Token payment model.

The main products attracting traffic are three: Phanthy Pantheon, upgraded from Paradigm’s model development platform; PhanRouter, a unified gateway for large model APIs; and PhanClaw, responsible for Token security and permission control.

The core is Phanthy Pantheon, mainly for B-end Token payments, also supporting C-end operations, deeply integrating cloud services from PhanthyCloud, over 30,000 models adapted via ModelHub XC, and covering over 100 million consumer electronics terminals.

The core idea of Pantheon is to combine generative AI with a systematic world model.

If compared to a hospital, generative AI is like registration and triage, understanding needs and coordinating processes; the world model is like a specialist doctor, responsible for analysis and decision-making. Together, they turn complex tasks into “deliverables.”

Based on products like Pantheon, monetization of API services relies on Token “pay-per-use,” with revenue still limited—only 79.9 million yuan in 2025, accounting for just 1.1%, but with a growth rate of 129.2%.

The explosion of Token consumption driven by intelligent agents makes the API sector highly flexible; Dai Wenyuan’s Token revenue for January-March 2026 alone has already exceeded the total for all of 2025.

In Dai Wenyuan’s design, if the AI platform is the base, API is the access point, and Agentic AI is the “spear” penetrating industry pain points.

This sector, Paradigm’s team builds agents to directly deliver business results, mainly adopting a “pay-for-results” model.

For example, in wind power and photovoltaic energy scenarios, it can help clients improve power generation revenue and reduce deviation assessment costs through precise output prediction and trading strategy optimization.

Dai Wenyuan hopes this sector can generate more “AI+X” applications, gradually expanding into multiple industry scenarios.

From “cloud” to “computation”, he realizes that collaboration is essential.

He led Paradigm’s investment in domestic GPU manufacturer Canaan, acquiring over 9% of shares; through the funds he invested in, he also invested in another domestic GPU company, Tianhesi Zhixin.

At the end of January, he appeared at Canaan’s launch event, jointly announcing the “one million tokens for one cent” inference cost plan with Canaan’s co-CEO Wang Zhan.

A new narrative

43-year-old Dai Wenyuan, with a prestigious university background and experience in large companies, has been a competitive player since childhood.

“I participated in many competitions from a young age… gradually training myself to be calm in the arena,” he said.

He deeply understands how to reflect and surpass.

At the end of February, Dai issued an announcement emphasizing that only 16.1% of revenue came from software services, “core main business is centered on computing power.” The enterprise service sector (4Paradigm), which he started, has faded into a sub-business.

He believes that the AGI (Artificial General Intelligence) architecture should be split into two parts: AI Agent and world model.

AI Agent handles human interaction, while the world model understands the rules of the world’s operation and solves final problems.

Dai Wenyuan believes that gathering all related capabilities will form the foundation for achieving AGI, with “AI Agent + world model” as the core path of technological evolution.

For example, his consumer electronics sector “Paradigm Navigation” focuses on the wearable device market, starting from end-side AI agents in a soft-hardware integrated approach.

It can provide clients with end-side AI general modules, with appearance design and details customizable by clients, while hardware configuration and AI algorithms are defined and optimized by Paradigm.

In 2025, the total shipment of end-side AI modules provided to Lenovo exceeded 1 million units.

At the end of January 2026, Dai Wenyuan announced that the brand system and strategic architecture had been upgraded. With this grand new narrative, he is also returning to A-shares.

In late March, Paradigm Intelligence announced that it had completed listing counseling registration, planning to go public on the Shenzhen Stock Exchange. The Hong Kong stock prospectus shows that in September 2023, Dai Wenyuan and institutions like Boyu Jingtai signed a bet agreement.

The agreement stipulates that Paradigm must complete an A-share listing before September 28, 2026, with a valuation of no less than 7.5 billion USD, roughly three times its current market value, and the A-share issuance scale must be at least 10% of the total post-issuance share capital.

If not achieved, investors have the right to require Dai personally to buy back his domestic shares, without using company funds.

The window is closing; telling a good AI story, time waits for no one.

Image source: Paradigm Intelligence, except where noted