The archive · Hardware & Devices · Product decision · 2022–2026
Ropet bets a silent desktop AI pet, not a talking chatbot, is China's next hit
Ex-ByteDance designer He Jiabin drops walking and voice to sell a ¥2,000 desk 'future pet'; ~20,000 shipped, <4% returns, >$10M Series A
Ropet (Mengyou Intelligence / 萌友智能)
What the business is
Ropet makes a ¥2,000 desktop AI pet robot without voice or dialogue: it senses touch and presence, learns the owner's routines, and reacts with eyes, sounds, and movements.
Starting capital：Pre-A ¥10M (RMB) from Zhenzhi VC and Lenovo Zhixiang in late 2024; Series A over US$10M in two closings (Sept and Nov 2025), announced 2026-04-09
How it started
Mengyou Intelligence (萌友智能) was founded in 2022, incubated by Zhenzhi VC (真知创投). Co-founder He Jiabin had led Luka, a children's reading robot, and worked at Microsoft Research Asia, Baidu, and ByteDance; co-founder Zhou Yushu holds a Stanford PhD in HCI. The first project — a complex embodied robot — struggled to raise money, so in June 2024 He took over as CEO and pivoted to a much lighter desktop device, cutting walking capability and related R&D.
What happened
A demo prototype was ready in four months and won a ¥10M Pre-A round from existing shareholders Zhenzhi and Lenovo Zhixiang. The team spent it on a CES 2025 debut and two crowdfunding campaigns (Kickstarter late 2024 and Japan's Makuake in Aug 2025), which together drew 4,063 backers and ¥5.65M (RMB), then reached mass production in Sept 2025. By April 2026 cumulative shipments neared 20,000 units, 70% of sales came from overseas, the return rate stayed below 4% versus an industry norm around 30%, 90-day retention ran 80–90%, and accessory repurchase ran 70–80%. In March 2026 Ropet announced a Universal Pictures 'How to Train Your Dragon' co-brand, with presales passing 1,000 units in five days.
How it ended up
Still scaling: as of April 2026 Ropet had closed a Series A over US$10M, shipped ~20,000 units since Sept 2025, and was expanding beyond JD/Tmall and its overseas site into offline tech-experience stores, while building a 'pet brain' model that learns human-pet interaction patterns.
Background
Ropet (Mengyou Intelligence, 萌友智能) is a Beijing startup selling a ¥2,000 desktop AI pet that deliberately does not talk. Founded in 2022 and incubated by Zhenzhi VC (真知创投), it spent its first years on a complex embodied robot that investors would not fund; in June 2024 co-founder He Jiabin became CEO and pivoted the team to a lightweight desk pet with expressive eyes, touch sensors, and no dialogue.
The product bet was 'weak companionship': a creature that tracks faces, responds to touch, remembers routines, and expresses emotions, instead of answering questions. Validation came overseas first — CES 2025, a Kickstarter campaign in December 2024 that raised about $400k and was named among the year's best AI consumer hardware, and Japan's Makuake in August 2025, which passed ¥27M in its first hour.
From September 2025 mass delivery onward, Ropet reported roughly 20,000 units shipped, 70% of sales outside China, a return rate below 4% (industry average near 30%), 90-day retention of 80–90%, and 70–80% accessory repurchase. In April 2026 it announced a Series A over US$10M split across September and November 2025 closings, led by the Beijing AI Industry Investment Fund with Frees Fund, Inno Angel, and Guotai Haitong, and a March 2026 How to Train Your Dragon co-brand that presold over 1,000 units in five days.
What has to be true
- The pivot removed walking and voice, cutting cost and complexity so a 4-month demo and ¥2,000 price point became possible.
- Overseas-first validation (CES, Kickstarter, Makuake) produced demand signals and cash before China channel spend.
- Refusing chat features protected a clear user promise — a pet for people who cannot keep one — instead of becoming another smart speaker.
- Retention and repurchase data (80–90% 90-day retention, 70–80% accessory repurchase) showed the emotional loop closed before the company scaled distribution.
What can be applied
When a category's consensus is 'AI toy equals LLM plus shell', the winning edge can be subtraction: pick the narrow emotional job, refuse feature creep, and prove demand abroad before scaling at home.
Aftermath
As of April 2026, Ropet said it was 'the fastest to commercialization in the track': ~20,000 units shipped since Sept 2025 with under 4% returns, sales on JD, Tmall, and its overseas site plus offline tech-experience stores, and a Universal How to Train Your Dragon line in presale. The next stated step is a 'pet brain' decision model that learns from real human-pet interaction data, with future pet variants planned — while Chinese rivals such as Yukai-backed Mirumi and other emotion-companion startups entered the same niche in 2026.
Sources
- AI机器萌宠企业「Ropet萌友智能」完成数千万A1轮融资 | 36氪首发
- 出货量近2万台、退货率低于4% 萌友智能完成超千万美元A轮融资
- AI机器宠物ropet销量近2万台，萌友智能完成超千万美金A轮融资
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card