Wangsu's co-lead ties Sand.ai's video models to a listed company's network nodes for low-latency inference, and the funding backs a full MAGI-2 launch against Kling and PixVerse.
Sand.ai, a Chinese developer of AI video-generation models, has raised US$200 million in a Series A+ round, TechNode reported, ahead of the planned release of the full version of its MAGI-2 model. CoStone Capital (基石资本) and the Shenzhen-listed Wangsu Science & Technology jointly led the round, which Sand.ai announced on November 5. It is the company’s third financing of 2026 and brings the total raised in 10 months to about US$300 million.
The round makes Sand.ai a unicorn, a private company valued above US$1 billion, according to DealStreetAsia, which puts the valuation above that mark; Sand.ai did not disclose a valuation and the Chinese reports do not state one. Sand.ai said it will spend the money on video-model research and training, training and inference infrastructure, product development and international expansion.
The raise follows Sand.ai’s August release of MAGI-2 Preview as an open-source model. Reports describe the full MAGI-2 as a larger mixture-of-experts model, an architecture that activates only part of its parameters for each request, due "soon" without a date. For investors, the round shows how much capital a company with modest revenue can draw while Kuaishou’s Kling and other rivals raise far larger sums.
Wangsu’s stake links the funding to inference infrastructure
Wangsu’s role is the best-documented part of the deal. CLS reports that the company announced on September 29 that it would invest RMB300 million of its own funds in the A+ round, and that agreements were signed on October 16. Under the structure Wangsu disclosed, it first provides Sand.ai’s operating company with convertible bonds and receives warrants, then makes the equity investment.
The two companies also plan to work together. According to CLS, Wangsu will supply computing capacity, network nodes, scheduling and security for audio and video inference, including low-latency inference at the network edge, while Sand.ai provides video models and inference technology. New Sand.ai models and technical results are to be deployed first on Wangsu’s platform.
Wangsu’s announcement also disclosed Sand.ai’s financials. Sand.ai’s first-half 2026 revenue was RMB17.398 million against a net loss of about RMB152 million, according to Wangsu’s unaudited disclosure. The loss is about 8.7 times revenue.
Other investors include Times Capital, Panlin Capital, Orient Securities Capital, Tianchuang Capital together with two Shandong railway investment entities, and the Shanghai state-backed Pudong Venture Capital and Shanghai Guofang. Existing shareholders iMPC (formerly Matrix Partners China), HyT Capital (华业天成) and CAS Star added funds, per 36Kr, which republished a PEDaily report, and DealStreetAsia.
MAGI-2 Preview is the model on which the pitch rests
Sand.ai’s own site describes MAGI-2 Preview as a 114-billion-parameter model that generates video and audio together and activates about 6 billion parameters per token. CLS and 36Kr add that a distilled version generates a 10-second 1080p clip for roughly RMB0.5, a cost estimate based on renting eight Nvidia H100 chips by the month at the time. 36Kr, citing PEDaily, calls MAGI-2 Preview the first open-source mixture-of-experts video model at the 100-billion-parameter scale, with 114 billion parameters in total, a claim this story could not check independently.
The company also sells a product. VidMuse, a creator tool launched in January 2026, passed US$10 million in annual recurring revenue within two months, CLS reports. Its Pro plan costs US$39 a month, and it integrates third-party models from Kling, Seedance and Veo alongside Sand.ai’s own.
The founder is Cao Yue, a Tsinghua PhD who worked at Microsoft Research Asia, led a visual-model center at the Beijing Academy of Artificial Intelligence and co-authored the Swin Transformer paper, per 36Kr.
Larger rivals set the bar for the full MAGI-2 release
The competitive field is well funded. CLS names Kling, PixVerse and Shengshu as peers. It reports that Alibaba Cloud led a roughly RMB2 billion Series B for Shengshu in April, that PixVerse’s developer closed a Series C of RMB2.98 billion in July, and that Kling’s capital increase of RMB19.802 billion closed on October 30 with Kuaishou keeping control. 36Kr puts Kling’s post-money valuation at US$18 billion.
Against those numbers, a US$200 million round buys Sand.ai time rather than scale. Its edge, according to the reports, is the cost of generating video, backed by the open-source MAGI-2 Preview.
The release date and benchmark results for the full MAGI-2 are the next things to watch. Whether the model is also open-sourced, and whether Wangsu’s network begins serving it in production, will show if the round converts into usage.