Junjie Yan

AI · foundation models
闫俊杰IOChinaAge ~37

Founder, Chairman, CEO, and CTO of MiniMax · Former Vice President of SenseTime

From a county-born math student to an AI PhD and SenseTime research executive, he bet on AGI before ChatGPT at 32, scaled multimodal and global consumer products, and took MiniMax to the Hong Kong exchange four years later.

Started after workingBig-tech executive turned founderScaled company$100M+ · 2026-06PublicChina to globalFrom Mainland ChinaBorn 1989Top-school trainedBig-tech backgroundAcademic background

Key numbers

Each figure includes its definition, date, and evidence level.

1989Born · age ~37
~32First venture · 2021
Financial filing$116.6M↗First-half 2026 revenue · 2026-06
Media reportAbout HK$4.82B ($619M)↗Hong Kong IPO proceeds · 2026-01
Financial filing27.638M↗Average monthly active users in the first nine months · 2025-09
Financial filing1.774M↗Paying users in the first nine months · 2025-09
Financial filing$73.9M (63.4%)↗Enterprise and open-platform revenue · 2026-06
Financial filing17.9% / -$293.0M↗Overall gross margin / adjusted net loss · 2026-06

Key stages

What he was doing at each age

~17

A county student enters mathematical training

~32

Left a major tech company to start a business at 32

~34

Did not abandon 2C after the takedown

~35

Multiple products go global

~37

Completed an IPO four years after founding

Geographic moves

The city behind each change in direction

  1. Xiayi, Henan
    1989 Born in Xiayi County, Shangqiu, Henan
  2. Nanjing
    ~17 A county student enters mathematical training
  3. Beijing
    ~21 PhD student in pattern recognition and artificial intelligence~25 Research intern~26 AI researcher → industrial R&D professional
  4. SenseTime
  5. Shanghai
    ~32 Left a major tech company to start a business at 32
  6. China
    ~33 Founder, CEO, and CTO
  7. global markets
    ~34 Did not abandon 2C after the takedown
  8. Global markets
    ~35 Multiple products go global~36 Chairman, CEO, and CTO
  9. Shanghai
  10. Hong Kong public market
    ~37 Completed an IPO four years after founding
  11. Global markets
    ~37 Chairman, CEO, and CTO of a listed company

Selected work

A multimodal model platform for developers and enterprises, offering usage-based APIs, Token Plan, dedicated resource pools, and model licenses

Talkie

An AI character and companion product for overseas markets, corresponding to the domestic product Xingye

Hailuo AI

An AI video and multimodal creation product for global users

How the business works

Category
AI · foundation models / multimodal / AI applications
Users and payers
Two different transactions: individuals pay for subscriptions and in-app virtual goods on Talkie / Xingye, Hailuo AI, and Audio (some products also carry ads); developers and enterprises buy Open Platform
How customers pay
Consumer subscriptions + in-app purchases + ads; enterprise customers pay by token or generated resources, while large customers can purchase dedicated inference resource pools, custom services, and model licenses
Main costs
Inference cloud services accounted for 92.7% of cost of sales in the first nine months of 2025, plus platform commissions; model training, fine-tuning, and experiments are included in R&D rather than cost of sales
Gross-margin layers
Consumer products had a 4.7% gross margin in the same period, versus 69.4% for enterprise and open-platform business—the revenue mix determines gross margin, not scale
Evidence level
Financial-report figuresIn the first half of 2026, enterprise and open-platform revenue was $73.9 million, or 63.4%, up 703.1% year over year; overall gross margin was 17.9%, R&D was $296.9 million, and adjusted net loss was $293 million
Profit logic
Increase GPU utilization and lower unit token costs while directing traffic toward higher-margin APIs and enterprise services
Main risk
The training race can still consume all gross profit—company-level profitability remains unproven

MiniMax's consumer products and enterprise platform are two different transactions. Talkie / Xingye, Hailuo AI, and Audio sell subscriptions and in-app virtual goods to individuals, and monetize through ads in some products. In the first nine months of 2025, consumer-product revenue was $38.02 million, or 71.1% of total revenue; Talkie / Xingye contributed $18.75 million and Hailuo AI $17.464 million. Prospectus

Open Platform charges developers and enterprises by token or generated resources; large customers can also purchase dedicated inference resource pools, custom services, and model licenses. As of September 2026, the platform offered usage-based APIs, Token Plan, prepaid credits, and video/voice resource packages. Official pricing In the first half of 2026, this business generated $73.9 million, or 63.4% of revenue, up 703.1% year over year, making it the main revenue source. Interim results

The cost evidence is also clear: the prospectus shows that inference cloud services accounted for 92.7% of cost of sales in the first nine months of 2025, plus platform commissions; model training, fine-tuning, and experiments were included in R&D rather than cost of sales. Consumer products had only a 4.7% gross margin in the same period, versus 69.4% for enterprise and open-platform business. In the first half of 2026, overall gross margin was 17.9%, but R&D expenses reached $296.9 million and adjusted net loss was $293 million.

So the business's profit logic is to increase GPU utilization and lower unit token costs while shifting more traffic toward higher-margin APIs and enterprise services. The financial results show improving transaction scale and gross margin, but also show that the training race can still consume all gross profit; MiniMax has not yet demonstrated company-level profitability.

Timeline

  1. Born in Xiayi County, Shangqiu, Henan

    Starting outChina · Xiayi, Henan

    The birthplace comes from local and profile reports and was not disclosed in the company's prospectus; only the birth year, 1989, is confirmed, so all ages are approximate.

    Source ↗
  2. Entered Southeast University to study mathematics

    About 17 years oldStarting outXiayi, Henan → Nanjing · Undergraduate

    He followed a county education path into a science and engineering university, building the mathematical foundation for his later shift to pattern recognition and artificial intelligence.

    Source ↗
  3. Graduated and entered the Institute of Automation, Chinese Academy of Sciences, for a PhD in AI

    About 21 years oldTurning pointNanjing → Beijing · PhD student in pattern recognition and artificial intelligence

    He moved from general mathematical training into AI research and received his PhD in 2015.

    Source ↗
  4. Interned at Baidu Research during his PhD

    About 25 years oldTurning pointBeijing · Research intern

    He researched face detection and recognition and received a Baidu scholarship; the experience showed him the gap between academic results and industrial use.

    Source ↗
  5. Received his PhD, completed postdoctoral research at Tsinghua, and joined SenseTime

    About 26 years oldMilestoneBeijing · AI researcher → industrial R&D professional

    He then spent more than six years at SenseTime, rising to vice president and deputy dean of the research institute, and moving from individual researcher to technology manager.

    Source ↗
  6. Left SenseTime and founded MiniMax in Shanghai

    About 32 years oldTurning pointSenseTime → Shanghai · Founder, Chairman, CEO, and CTO

    He bet on AGI and multimodality before ChatGPT was released, moving from an executive role in a mature organization to the top role responsible for technology, operations, and the board.

    Source ↗
  7. Released its first large models and AI companion product Glow

    About 33 years oldMilestoneChina · Founder, CEO, and CTO

    From day one, the team bet on multimodal capabilities across text, voice, and vision, testing model value directly through 2C products.

    Source ↗
  8. Glow was taken down over regulatory licensing issues, and the team adjusted its domestic and overseas product paths

    About 34 years oldFailureChina → global markets · Founder, CEO, and CTO

    After approval, the product returned as Xingye, while Talkie expanded overseas. The team kept its conviction that models must be used by ordinary people, but changed the product and geographic distribution.

    Source ↗
  9. Built a portfolio of Talkie, Xingye, Hailuo AI, and an open platform

    About 35 years oldMilestoneGlobal markets · Founder, CEO, and CTO

    The company completed a large funding round, with media reporting a valuation of about $2.5 billion; a valuation is a financing price, not a transaction price or personal wealth.

    Source ↗
  10. The prospectus disclosed 27.638 million average monthly active users and 1.774 million paying users

    About 36 years oldMilestoneGlobal markets · Chairman, CEO, and CTO

    It had served about 212 million users measured by accounts or devices, with revenue of $53.437 million in the first nine months of the year; one person may correspond to multiple accounts, so cumulative accounts cannot be treated as deduplicated users.

    Source ↗
  11. MiniMax listed on the Hong Kong Stock Exchange, raising about HK$4.82 billion

    About 37 years oldMilestoneShanghai → Hong Kong public market · Chairman, CEO, and CTO of a listed company

    The first-day closing price was about 109% above the offer price, putting the company's market capitalization at about $13.7 billion; proceeds belong to the company, and market capitalization is not Junjie Yan's personally realizable wealth.

    Source ↗
  12. First-half revenue reached $116.6 million

    About 37 years oldMilestoneGlobal markets · Chairman, CEO, and CTO of a listed company

    Revenue grew 283.1% year over year, with the open platform and enterprise services accounting for 63.4%; adjusted net loss was $293 million in the same period, so growth still rested on heavy investment. The company already had revenue, so this milestone is not labeled “first time making money.”

    Source ↗

From Mathematics to Industrial AI

Junjie Yan went from Xiayi County in Henan to study mathematics at Southeast University, then transferred in 2010 to the Institute of Automation at the Chinese Academy of Sciences for a PhD in artificial intelligence. An internship at Baidu Research during the PhD was his first chance to test face-recognition research in a real industrial setting. After receiving his PhD in 2015 and completing a postdoctoral stage at Tsinghua, he joined SenseTime.

During more than six years at SenseTime, he grew from researcher to vice president and deputy dean of the research institute. More exaggerated CTO titles have circulated in public materials, but MiniMax's prospectus does not fully support those claims, so this entry uses only the positions confirmed by the prospectus. The important point is not how impressive the title sounds, but that he gained experience managing large teams, compute, and commercial projects.

Founding Before ChatGPT

In December 2021, he left SenseTime to found MiniMax, serving simultaneously as chairman, CEO, and CTO. This was before ChatGPT's release, showing that the team was betting on generative models and multimodal capability itself rather than pivoting after the trend appeared.

MiniMax tested whether people would use models early through products such as Glow. In 2023 Glow was taken down over domestic generative-AI regulatory licensing issues; after approval it returned as Xingye, while Talkie continued expanding overseas. This setback says more about the team's basic judgment than funding news: it did not abandon the consumer direction, but could change product form and market boundaries.

Multiple Products, Globalization, and the Public Market

In 2024, MiniMax formed a portfolio of Talkie, Xingye, Hailuo AI, and an open platform. By September 2025, the prospectus disclosed average monthly active users of about 27.638 million over the first nine months, about 1.774 million paying users, and revenue of $53.437 million. The cumulative figure of 212 million is an account or device measure, not deduplicated people.

In January 2026, MiniMax listed in Hong Kong, completing the transition from private startup to public company in about four years. Its first post-listing interim results had more operating meaning: first-half revenue was $116.6 million, with the open platform and enterprise services representing 63.4%, suggesting that revenue structure was moving from early entertainment and companion products toward APIs and enterprise use.

The same results provide the other side that must be viewed alongside it: adjusted net loss of $293 million. Junjie Yan has shown that the team can expand quickly across models, products, and capital markets; the next question is whether this system can keep growing at a more sustainable cost.

In their words
AGI is not a killer app; it is a product ordinary people use every day.
—— Junjie Yan ↗

More quotes

When no one uses a product, or when it does not make money, you certainly cannot blame the users.
2024#Commercialization responsibilitySource ↗
Intelligence can scale almost without limit; energy and compute cannot.
2026#Compute efficiencySource ↗

My read

Junjie Yan is easiest to write as a scholarly genius who guessed ChatGPT early. The more accurate path is that he first carried AI-PhD training into Baidu and SenseTime, spent six years learning to manage large-scale R&D and commercialization, and only then became the head of a startup during a non-consensus period. MiniMax's starting point was not only a model judgment, but also the dispatching capability accumulated in a mature organization.

The first lesson worth copying is not treating advanced technology as a reason users must pay. After Glow was taken down, the team did not blame regulators or users; it returned as Xingye while Talkie continued validating consumer products overseas. Second, it obtained feedback directly from companion, video, and voice products, then made the open platform and enterprise services a larger source of revenue. Third, concentrating CEO and CTO authority lets the technical route and product commercialization move quickly together, but public markets also require more sustained disclosure and accountability.

In business terms, MiniMax's key is not the number of products, but two forms of monetization from the same base model: consumers pay subscriptions or top-ups for companion, video, and voice products, while developers and enterprises pay for APIs, dedicated resource pools, and model licenses. In the first half of 2026, enterprise and open-platform business already represented 63.4% of revenue, showing that the commercial center was moving from early consumer products toward production calls.

The caution is that this is not an “overseas expansion template” an ordinary small team can easily copy. The SenseTime executive experience, elite research team, enormous compute investment, and capital-market window all formed the starting point. First-half 2026 revenue of $116.6 million and a 17.9% gross margin are hard figures, but adjusted net loss in the same period was $293 million; IPO, market capitalization, and user scale cannot replace long-term profitability.

Next founder