China's Sand.ai Becomes a Unicorn With a $200 Million Round as Its Open-Source MAGI Video Models Take On Kling and Seedance

A Chinese artificial intelligence startup building some of the world's most advanced open-source video generation models has joined the ranks of the country's unicorns, as investors bet that AI-generated video will be one of the next big frontiers of generative AI.

Sand.ai announced a $200 million Series A+ funding round on Thursday, November 5, according to TechNode, bringing its total fundraising in 2026 to about $300 million across three rounds in roughly 10 months, according to 36Kr. The round gave the company "unicorn status," a valuation of at least $1 billion, according to TechNode, although the exact valuation was not disclosed.

The round was co-led by Jiyuan Capital, also reported as Cornerstone Capital, and Wangsu Science & Technology, a listed Chinese content delivery network company. Wangsu invested 300 million yuan for about 2.1 million Series A+ preferred shares, according to 36Kr.

Other participants included Times Capital, Panlin Capital, Dongzheng Capital, Tianchuang Capital, Shandong Railway Fund, Pudong Venture Capital and Shanghai Guofang, while existing investors Matrix Partners China, Huaye Tiancheng and CAS Star also returned.

A founder from China's AI elite

Sand.ai was founded in 2024 by Cao Yue, a computer science PhD from Tsinghua University, according to 36Kr. Cao is a co-author of Swin Transformer, an influential computer vision architecture, and won the Marr Prize at the International Conference on Computer Vision in 2021, one of the field's most prestigious awards.

He previously worked as a principal researcher at Microsoft Research Asia, the US technology giant's Beijing lab, which has produced many of China's leading AI scientists, and led the Vision Model Center at the Beijing Academy of Artificial Intelligence.

The MAGI models

Sand.ai's core technology is its MAGI family of video generation models. Unlike most video models at the time of its founding, which used diffusion techniques, Sand.ai chose an autoregressive approach, generating video sequentially, according to 36Kr. It later adopted a mixture-of-experts (MoE) architecture, in which only parts of the model are activated for each task, improving efficiency.

In August, the company released and open-sourced MAGI-2 Preview, with 114 billion total parameters, of which about 6 billion are activated for each run. 36Kr described it as "the world's first 100-billion-level open-source MoE video generation model." A distilled version can reportedly generate a 10-second 1080p video for about 0.5 yuan, roughly one-tenth the per-second cost of mainstream models.

Sand.ai plans to release the full version of MAGI-2 soon, according to TechNode, with a larger parameter scale.

Models and products

Sand.ai has paired its models with consumer products. Its VidMuse app, an AI video creation agent for content creators, exceeded $10 million in annual recurring revenue about two months after launching at the beginning of the year, according to 36Kr.

"In the video model direction, it is difficult to only be a company that sells APIs. Models and products must be driven by two wheels," Cao said, according to 36Kr.

The company plans to use the new funding for video model research, training and inference infrastructure, product development and international expansion, according to TechNode.

The video generation race

AI video generation has become one of the most competitive areas of generative AI. In China, Kuaishou's Kling and ByteDance's Seedance are among the leading video models, while Alibaba, Tencent and MiniMax have also released video tools. Globally, OpenAI's Sora, Google's Veo and startups such as Runway and Luma compete for users.

Video generation is technically demanding and expensive, requiring vast computing power to train and run models. Sand.ai's focus on efficiency through MoE architecture and its low per-video cost could give it an advantage, particularly as Chinese companies face restrictions on access to the most advanced US AI chips.

China's open-source strategy

Sand.ai's decision to open-source its models reflects a broader trend in China's AI industry. Companies such as DeepSeek, Alibaba's Qwen team and others have released powerful open-source models, building global developer communities and challenging the closed models of US companies. Open-source releases can accelerate adoption and attract talent, while companies make money from products and services built on their models.

A hot market for Chinese AI

Sand.ai's round was one of several large Chinese technology deals announced on November 5. Semiconductor equipment maker Bangxin Semiconductor raised more than 600 million yuan, and photonic chip company Huayi Yingfei Photonics raised several hundred million yuan, according to Tech Startups. The deals reflect strong investor appetite for Chinese AI and deep technology companies, supported by government and state-linked capital.

What it means for creators

AI video tools are transforming content creation, allowing creators to produce videos, animations and effects that once required professional studios. For India's huge creator economy, which includes millions of YouTubers, Instagram creators and short-video makers, as well as a large film and advertising industry, low-cost AI video tools could lower production costs and open new creative possibilities.

At the same time, AI video raises concerns about deepfakes, misinformation and copyright. Governments, including India's, are developing rules to label AI-generated content and prevent misuse.

Funding history

Sand.ai's rapid fundraising reflects strong investor interest in China's AI sector. Before the latest round, the company raised more than $100 million across two rounds in June, with investors including IDG, Baidu Ventures, Sinovation Ventures, Capital Today and Source Code Capital, according to 36Kr. The participation of so many well-known Chinese investors underlines the company's standing in the market.

The involvement of Wangsu Science & Technology, which operates content delivery networks, is strategically significant, as efficient delivery of video content is crucial for AI video applications.

Sand.ai's approach of releasing open-source models while building paid products resembles strategies used by other AI companies, which use open models to build developer communities and brand recognition, then generate revenue from applications, enterprise services and cloud access. With video generation still early in its commercial development, the company is betting that efficiency and openness will help it scale faster than larger, better-funded rivals.

Investors will also watch how Sand.ai navigates competition from Chinese technology giants, which have far greater computing resources and distribution through their existing apps and platforms.

What comes next

Sand.ai will focus on releasing the full MAGI-2 model, scaling VidMuse and expanding internationally. Its progress will test whether a young startup can compete with China's internet giants in video AI, and whether its open-source, efficiency-focused approach can win in a market where computing costs are a major barrier.