--- title: "Chinese video-model startup Sand.ai says a $200M round lifts it past the unicorn line" date: 2026-11-07 category: Capital & Markets site: NeuroAI canonical: https://neuroai.site/a/na-capital-sandai-video-model-200m language: en --- # Chinese video-model startup Sand.ai says a $200M round lifts it past the unicorn line > Sand.ai (三呆科技), a video-generation large-model (大模型) company, closed a US$200M A+ round in early November 2026 — its third raise this year — taking cumulative 2026 funding to US$300M, with the company saying it had crossed the unicorn line. A founder with a background in open-source chat models is now betting that the next mass-market AI surface is video. On 5 November 2026, Sand.ai (三呆科技) said it had closed a US$200M A+ round, its third financing of the year, and that the money lifted it over the unicorn threshold. The signal is not the headline number alone. It is where the capital is flowing inside China's large-model (大模型) sector: away from general chat bots and toward specialized, media-native models with a clear path to product revenue. ## Key takeaways - Sand.ai (三呆科技) closed a **US$200M A+ round** (≈ RMB 1.4B) on 5 November 2026, its third raise of the year. - Cumulative 2026 financing reached **US$300M** (≈ RMB 2.1B); the company said it has crossed the unicorn line. - The round was led by Cornerstone Capital (基石资本) and Wangsu Technology (网宿科技), with Shanghai state-linked investors and several returning backers. - Sand.ai builds **video-generation foundation models (视频生成大模型)**: it has open-sourced the MAGI series and ships a product called VidMuse. - The raise extends a clear trend — capital concentrating in media-generation and inference-ready models rather than yet another general assistant. ## From MOSS to moving pictures The company's lineage matters. According to the company, Sand.ai was founded in January 2024 by Cao Yue (曹越), whose background it describes as including Tsinghua, Microsoft Research Asia, the earlier AI venture Light Year (光年之外), and Fudan's MOSS team. The company frames that background as the reason it targeted video from the start. It open-sourced the MAGI family of video models and launched VidMuse, a video-generation product. The competitive field is crowded and well-funded. Kuaishou's Kling (可灵) and ByteDance's Seedance (即梦) iterated through 2026, each pushing on controllability, inference efficiency, and commercial fit. Sand.ai's bet is that an independent, open-weights video lab can still carve out enterprise and creator demand by pairing model quality with a usable product surface. ## Why investors paid up now Three forces converged. First, video is the most token-hungry modality in generative AI (生成式AI / AIGC); every minute of high-fidelity video consumes far more compute than text, so the economics reward teams that can compress cost per generated second. Second, China's short-drama (短剧) and advertising pipelines are hungry for on-demand video at scale, giving domestic video models a nearer-term commercial pull than many text-only rivals. Third, the 2026 financing environment favored companies that could show repeated, oversubscribed rounds — Sand.ai's three raises in ten months read as momentum, not distress. The cap table reflects that. The A+ was co-led by Cornerstone Capital (基石资本) and Wangsu Technology (网宿科技), a listed CDN and edge-computing firm that brings distribution reach. Shanghai state-linked vehicles — Pudong Venture Capital (浦东创投) and Shanghai Guofang (上海国方) — joined, alongside returning investors including Matrix Partners China (经纬创投), Hua Ye Tian Cheng (华业天成) and CAS Star (中科创星). The presence of both strategic infrastructure players and state capital is the 2026 template for a Chinese AI media company that wants to scale compute and reach together. ## What the money is for Sand.ai said proceeds go to video foundation-model R&D, training and inference infrastructure, product iteration, and overseas expansion. That infrastructure line is the real story: a US$200M round in this market is less about headcount and more about securing GPU time, storage, and the inference fleet needed to serve video at production volume. For a media-model company, the cost curve of serving requests — not the cost of training one flagship model — is the constraint that decides whether the unit economics ever work. According to Li An, chief scientist at BrainNet (脑机网), the clustering of late-2026 rounds into video and inference-native models shows capital moving from "who has the biggest model" to "who can deliver tokens cheaply at scale" — a shift that rewards engineering discipline over parameter bragging. ## What readers can do now - **If you build with video AI, benchmark on cost-per-minute, not demo quality.** Sand.ai, Kling, and Seedance differ most on price and throughput once you move past showcase clips. - **Watch the open-weights versus hosted split.** A company that open-sources (MAGI) and sells a product (VidMuse) is hedging two revenue paths; that duality is worth tracking as a sustainability signal. - **For China-focused creators, trial the domestic options now.** Local video models are iterating monthly, and early-adopter terms from vendors chasing market share may not stay this favorable later. ## Honest limitations This article relies on two Chinese financial-news accounts of the financing — Paper.cn (澎湃新闻) and a STAR Market Daily (科创板日报) dispatch carried by Tencent News — plus Sand.ai's own statements. The US$200M A+ size, the US$300M ten-month cumulative, the unicorn claim, and the list of lead and participating investors are reported as stated and not independently audited here. I did not obtain the term sheet or a prospectus, so valuation, post-money structure, and any secondary components are not verified. The competitive comparison to Kling and Seedance is framed by public product releases rather than side-by-side benchmarks. I have not covered Sand.ai's technical architecture in depth, and the Li An remark is a qualitative trend observation, not a sourced quantitative claim. RMB figures are approximate conversions from USD at roughly 7.1. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-capital-sandai-video-model-200m Free to quote with attribution and a link to the original.