🌏 中文版
Funding Details
| Item | Value |
|---|---|
| Company | Sand.ai (Beijing, China) |
| Round | Series A+ |
| Amount | $200M |
| Lead investors | CoStone Capital, Wangsu Science & Technology (Shenzhen-listed) |
| Participants | Multiple Chinese venture firms and state-backed institutions, plus existing shareholders iMPC (formerly Matrix Partners China), CAS Star, and HyT Capital re-upping |
| Valuation | Crossed $1B (first time reaching unicorn status; exact prior-round valuation undisclosed) |
| Total funding | $300M (across three rounds in the past 10 months) |
| Founded | 2024 |
| Employees | Fewer than 30 as of April 2026 (per apiyi reporting; current headcount after this round undisclosed) |
What the Company Does
Sand.ai builds foundation models for video generation. It was founded in early 2024 by Cao Yue, a Tsinghua PhD who co-authored the Swin Transformer paper (winner of ICCV's Marr Prize) and previously worked at Microsoft Research Asia and the Beijing Academy of Artificial Intelligence, before co-founding Light Year Beyond as algorithm co-founder in 2023.
Its core product is the MAGI family of autoregressive video models, released fully open source under Apache 2.0 — a different path from Stability AI (image-focused) or Midjourney (closed, subscription-based). The latest MAGI-2 Preview is a 114-billion-parameter mixture-of-experts model that jointly generates video and matching audio, and ranked sixth globally on the Artificial Analysis image-to-video leaderboard at launch. Its commercial product, VidMuse, is a paid video-creation service for music videos, ads, and animated content; the company says VidMuse crossed $10M in annualized recurring revenue within two months of its January 2026 launch — a self-reported figure that hasn't been independently audited.
What This Round Signals
For the Agent Ecosystem
This round follows Kuaishou's Kling AI unit closing roughly $2.9B (about RMB 19.8B) in October — China's generative-video sector is seeing back-to-back mega-rounds, and the center of gravity has shifted from "whose model performs better" to "whose training and inference infrastructure can keep up." One of the lead investors, Wangsu Science & Technology, is a CDN/edge-computing company. Per a September 29 filing with the Shenzhen Stock Exchange, it structured its roughly RMB 300M (about $44.5M) investment as a convertible note plus warrants for about 4.4% equity, paired with an agreement to collaborate on audio/video inference, low-latency edge inference, and enterprise deployment — meaning part of this round is a vertical-integration play by an infrastructure provider, not pure growth equity.
What Investors Are Betting On
Wangsu's disclosed logic is direct: lock in priority deployment rights for Sand.ai's future models on its edge network, effectively signing up a model supplier before a competitor does. CoStone Capital co-leading alongside a strategic infrastructure investor shows generalist growth capital still sees upside in vertical video models. State-backed participants — Shanghai Pudong Venture Capital and Shanghai Guofang among them — also signal local government capital continuing to back China's leading video-gen players amid chip and compute constraints.
Numbers Worth Watching
- Three rounds in 10 months totaling $300M averages roughly $100M per round — an unusually dense fundraising cadence, suggesting investors still see room to stake positions in a sector that hasn't produced a clear winner yet.
- VidMuse reportedly crossed $10M ARR within two months, against a team of fewer than 30 people as of April — if both figures hold, the revenue-per-employee number is striking, but they're from different points in time and shouldn't be divided together as a conclusion, just tracked as a signal.
- Against Kuaishou's Kling raising roughly $2.9B in a single round, Sand.ai's $300M over 10 months is still a smaller league — this unicorn milestone looks more like clearing an entry threshold than securing a leading position.
Watchlist Status
Sand.ai is not currently on the watchlist. Recommend adding it to section C6 (Multimodal AI Generation Tools), with a secondary F1 (China AI Ecosystem) regional tag, tracking the open-source MAGI model family, VidMuse's commercialization, and progress on the Wangsu edge-inference partnership.
Today's Takeaway
Wangsu didn't just write an equity check — it used a convertible note plus warrants to buy priority deployment rights instead. That tells you infrastructure providers now care more about locking in whose inference traffic they'll carry next than about financial returns alone. Edge-computing vendors are using investment contracts to pre-secure future compute customers rather than waiting for customers to show up — a different calculus than a traditional VC watching valuation multiples.