Recent News

2026/11/05

your location: Home> News Updates> Detail page
HyT Capital Portfolio | Sand.ai Raises US$200 Million in Series A+ Funding to Advance Video Generation

Sand.ai, an early portfolio company of HyT Capital that develops video generation foundation models and products, has raised US$200 million in Series A+ funding. We first invested in Sand.ai in its Pre-A+ round in 2024 and have continued to invest in subsequent rounds.

Sand.ai began by exploring autoregressive video generation, an approach that generates video segments in chronological order. The team has continued to develop its own model architectures and release its work openly. This round marks a milestone in that long-term effort.

On November 5, 2026, Sand.ai announced the close of the round, its third financing of the year. The company raised a total of US$300 million over 10 months. Its valuation rose roughly fivefold from the start of the year, and its post-money valuation exceeded US$1 billion, giving the company unicorn status.

The round brought together strategic investors, RMB- and US dollar-denominated funds, and Shanghai state-backed investors. CoStone Capital and Wangsu Science & Technology (300017.SZ) co-led the round. Jeneration Capital, Tianjin Venture Capital Management, Hengxu Capital, Panlin Capital and Orient Securities Capital Investment also participated, with Shandong Railway Investment Holding Group participating alongside Tianjin Venture Capital Management. Shanghai Pudong Venture Capital and Shanghai Guofang joined the round, while existing investors MPCi, HyT Capital and CASSTAR made follow-on investments.

The proceeds will primarily fund research and development for video foundation models, infrastructure for training and inference, further product development, and global market expansion. At the time of the announcement, Sand.ai said it planned to release the full version of MAGI-2, its next-generation video generation foundation model, with a larger mixture-of-experts (MoE) architecture.

From Autoregressive Video Generation to a 100-Billion-Parameter MoE Model

Since its founding, Sand.ai has focused on video generation foundation models and products. Its research has progressed from autoregressive generation and joint audio-video modeling to MoE models at the 100-billion-parameter scale, with advances in model capabilities and inference efficiency.

In April 2025, Sand.ai released MAGI-1, an open-source autoregressive video generation model. MAGI-1 generates video segments in chronological order, allowing clips to be extended continuously into longer sequences. In March 2026, Sand.ai and SII-GAIR released MagiHuman, which uses a unified single-stream Transformer to jointly model text, video and audio. Sand.ai also developed infrastructure such as MagiAttention and MagiCompiler to improve training and inference efficiency for large video models.

In August 2026, Sand.ai released MAGI-2 Preview with open weights, marking the launch of the world’s first open-weight MoE video generation model at the 100-billion-parameter scale. Its ultra-fine-grained MoE architecture has approximately 114 billion parameters in total and activates about 6 billion per token, allowing the model to increase capacity while limiting inference compute. Based on GPU rental rates at the time, Sand.ai estimated that the distilled model could generate a 10-second 1080p video at an inference cost of approximately RMB 0.50. At launch, MAGI-2 Preview ranked sixth on Artificial Analysis’s image-to-video generation leaderboard.

Alongside its foundation models, Sand.ai launched VidMuse, a video generation product that explores how generative AI can be used to create and consume video content. The company reported that VidMuse surpassed US$10 million in annual recurring revenue (ARR) within two months of launch.

With the new funding, Sand.ai will continue developing its next generation of large-scale video models and exploring how open-source foundation models and products can inform each other’s development. The company also plans to expand its global developer ecosystem and bring its technology to more commercial applications.

Investors on Sand.ai’s Long-Term Potential

CoStone Capital said, “Sand.ai has shown particular strength in developing large-scale video models, advancing the underlying technology and iterating quickly. We have long believed in the company’s technical direction and growth potential, and we believe it can become an independent video foundation model company with global reach.”

Wangsu Science & Technology, a provider of edge computing and security services, previously entered into a strategic partnership with Sand.ai. The two companies plan to collaborate on audio and video inference, low-latency edge inference, and industry applications to help deploy and commercialize video generation technology at scale.

“Sand.ai’s strengths in video foundation models complement our global edge computing and security capabilities,” Wangsu said. “Through our investment and business collaboration, we look forward to accelerating the adoption of high-quality video generation across industries.”

Shanghai Pudong Venture Capital said, “Pudong brings together research talent, technology development capabilities and opportunities for industrial applications. It is well positioned to connect international capital, technology and markets. Sand.ai has demonstrated a strong ability to innovate in video foundation models and the potential to compete globally. We see long-term potential in the company and look forward to seeing it build on the strengths of China’s AI ecosystem to make further progress in the global video generation market.”

Jeneration Capital said, “Video generation foundation models are moving from demos into production. Competition increasingly depends on overall model capabilities, inference efficiency and the strength of the developer ecosystem. Sand.ai has developed a distinctive approach to large-scale MoE architecture and engineering and is competing with leading global models through open-source releases. We believe Chinese teams building open-source video models can create lasting value, and we expect Sand.ai’s continued technical progress to earn it an important place in the global video generation market.”

MPCi, an existing investor, said, “We have invested in Sand.ai across four consecutive rounds, from its early technical exploration and the open-source release of MAGI-1 to the launch of MAGI-2 Preview. Over that time, we have seen the team make steady progress in video foundation models. The team has shown clear judgment about its technical direction, strong research and engineering capabilities, and an ability to iterate quickly. Each milestone has strengthened our long-term conviction. We will continue to follow new approaches in the field and look forward to Sand.ai’s next advances in the global video model market.”