Sand.ai raises 200 million yuan in A+ round

The round drew a mix of industrial capital, yuan and dollar funds, and state-backed investors, underscoring both the commercial and strategic interest in China’s AI stack. Basestone Capital and Nettrix Technology co-led the financing, while Shanghai-backed Pudong Venture Capital joined alongside Jeneration Capital, Guofu Innovation, Panlin Capital and East Money Securities-affiliated investors. Existing backers including Genesis Capital, Huaye Tiancheng and CAS Star also increased their stakes.

That composition is important. It suggests Sand.ai is not being funded purely as a speculative venture bet, but as part of a broader push to build domestic capacity in frontier AI models, especially in video. For Chinese investors, the appeal is a company that can pair open-source model releases with enterprise and creator-facing products. For state-linked investors, it aligns with the policy objective of strengthening local AI infrastructure and reducing dependence on foreign model ecosystems.

Sand.ai has tried to differentiate itself through architecture rather than marketing. The company said its next-generation MAGI-2 video foundation model, based on a larger mixture-of-experts design, will be released soon. It has already published MAGI-1, a autoregressive video model, and MagiHuman, a unified transformer for text, image and audio. In August, it open-sourced MAGI-2 Preview and said it was the first team to scale video generation to a 100-billion-parameter class model while releasing weights publicly.

The technical pitch is straightforward: more capable video generation without runaway inference costs. Sand.ai said a distilled version of MAGI-2 can generate 10 seconds of 1080p video at a cost of 0.5 yuan, a figure that, if sustainable at scale, could help close the gap between headline model quality and practical commercial deployment. In a market where compute bills often determine whether AI businesses can eventually generate margin, lower unit costs matter as much as model quality.

Investors appear to be betting that this combination of open-source credibility, engineering efficiency and productization can translate into revenue. Sand.ai’s VidMuse product reportedly surpassed a 10 million dollar annualized revenue run rate within two months of launch, a sign that demand exists beyond research labs. That offers a bull case for investors: if the company can convert model performance into a broader developer ecosystem and paid creative tools, it could become one of the few independent Chinese video-model platforms with both technical prestige and commercial traction.

The bear case is that the economics remain punishing. Video models are more expensive to train and serve than text models, competition is intensifying globally, and open-source leadership does not automatically translate into durable pricing power. Sand.ai will also face pressure to keep pace with rapidly evolving rivals in the U.S. and China, where foundation-model spending remains concentrated among the best-capitalized players.

For investors, the message is that China’s AI funding market is still willing to reward companies tackling the hardest part of generative AI, not just the fastest-to-monetize layer. The presence of industrial buyers, provincial capital and repeat venture backers suggests confidence that video AI is moving from novelty to infrastructure. The next test will be whether Sand.ai’s forthcoming MAGI-2 release can strengthen adoption enough to justify the valuation reset that comes with unicorn status.