THE ESSENTIALS IN 30 SECONDS

  • Sand.ai raised $200 million in a Series A+ round co-led by CoStone Capital and Wangsu Science & Technology, at a valuation reported above $1 billion.
  • The round follows two raises worth more than $100 million in June and brings the total to about $300 million in ten months.
  • Its MAGI-2 Preview, released as open source in August, is a mixture-of-experts video model with 114 billion parameters, about 6 billion active.

Sand.ai, a Beijing start-up that builds video-generation models, announced on November 5 that it had raised $200 million in a Series A+ round, an extension of its Series A. DealStreetAsia reports that the deal values the company at more than $1 billion. It is Sand.ai's third funding round of 2026, and the company has now raised about $300 million in ten months, after two rounds worth more than $100 million in total in June.

Who invested

According to DealStreetAsia, the round was co-led by CoStone Capital and Wangsu Science & Technology, a Shenzhen-listed company. A stock exchange notice cited by 36Kr says Wangsu planned to put 300 million yuan into the company in exchange for about 2.1 million Series A+ preferred shares, and to cooperate with it on audio and video inference and edge inference. Shanghai state-linked investors also took part, including Pudong Venture Capital and an investment arm of Shanghai International Group, and existing shareholders such as iMPC (formerly Matrix Partners China) and CAS Star added to their stakes. Earlier backers include IDG Capital, Baidu Ventures, Sinovation Ventures and Lollapalooza Capital, the family office of Meituan co-founder Wang Huiwen. Sand.ai plans to spend the money on training its video foundation models, inference infrastructure, product work and expansion abroad.

Models and products

Sand.ai was founded in early 2024 by Cao Yue, who holds a computer science PhD from Tsinghua University, co-authored the Swin Transformer paper and worked at Microsoft Research Asia. Its first model, Magi-1, generates video autoregressively. In August it released MAGI-2 Preview as open source. 36Kr describes it as a mixture-of-experts model with 114 billion parameters in total and about 6 billion active, and reports that a distilled version can produce a 10-second 1080p clip for about 0.5 yuan. The company has also built GAGA-1, which generates audio and video together, and sells VidMuse, a paid video-creation agent. Sand.ai says VidMuse passed $10 million in annual recurring revenue about two months after its launch in January.

What remains open

A final version of MAGI-2 is planned with a larger parameter count, but no release date has been given. 36Kr did not report a valuation figure. Neither source discloses user numbers or the license terms of the open model, so it is still hard to measure Sand.ai's position in a crowded Chinese video-generation market.