Sand.ai raises $200M Series A+ led by CoStone Capital and Wangsu, valuing video generator above $1B
The AMW Read
Known player with a new round; unicorn status and a CDN co-lead are segment-level signals, not a new entrant or a resolved debate, and the round is below the cross.§D threshold.
Sand.ai raises $200M Series A+ led by CoStone Capital and Wangsu, valuing video generator above $1B
Sand.ai announced on November 5 that it closed a $200M Series A+ round, co-led by CoStone Capital and content delivery network operator Wangsu (ChinaNetCenter), with Shanghai government-backed funds also participating. The round lifts the Chinese video generation startup's valuation above $1 billion, making it a unicorn. It is the company's third financing in 2026 so far, bringing cumulative funding to about $300M. Founded in January 2024 by Cao Yue, a former computer vision lead at Microsoft Research Asia, Sand.ai says the money will fund video foundation model research, training and inference infrastructure, product work and overseas expansion.
Sand.ai's distinguishing choice is an autoregressive architecture rather than the diffusion approach common in video generation. It open-sourced MAGI-1 in April 2025 and released MAGI-2 Preview in August 2026, a 114-billion-parameter mixture-of-experts model that ranked sixth in Artificial Analysis's image-to-video category at launch. The company claims a distilled version generates 10-second 1080p video for about RMB 0.5 (~¥12) in inference cost, a figure we have not verified. The Wangsu leg ties model development to edge distribution. We reported in October that Wangsu proposed a RMB 300M investment for a roughly 4.4% stake at an implied RMB 6.8B valuation, and the article now places that stake inside a co-led round. The article also reports that VidMuse, a music-to-music-video service, passed $10M in annual recurring revenue about two months after launch.
For builders, the open question is whether autoregressive video models can match diffusion pipelines on cost per second of output, and whether MAGI-1's open weights are a practical base while MAGI-2 remains in preview. For investors, the valuation rests more on research positioning and a distribution partnership than on disclosed revenue. The article gives ARR for one product and no figures for the core model business, usage, or the commercial terms of the Wangsu cooperation. The official MAGI-2 release, expected soon, is the next test of whether the preview's benchmark standing holds.
#VideoGeneration #GenerativeAI #SandAI #AutoregressiveModels #AIFunding #ChinaAI
