Sand.ai has raised $200 million at a valuation above $1 billion, becoming video AI’s newest unicorn. The Chinese startup announced on November 5 that it completed a Series A extension led by CoStone Capital and Wangsu Science & Technology, its third funding round of 2026. The round brings its total raised in the past ten months to $300 million, money it plans to spend on model research, computing infrastructure, product upgrades, and overseas expansion.
## The Round
The round drew a broad mix of investors, from industry capital and government funds to domestic yuan funds and US dollar investors. State-owned Pudong Venture Capital Group and an investment arm of Shanghai International Group joined, while existing investors including Matrix Partners China, CAS Star, and HyT Capital re-upped, according to a company announcement cited by DealStreetAsia.
Wangsu, the Shenzhen-listed cloud and network infrastructure provider, is more than a co-lead. It had already disclosed plans to invest 300 million yuan, about $45 million, in Sand.ai, through convertible bonds and warrants plus cooperation on computing, networking, and security for audio and video inference.
In unaudited figures in Wangsu’s filing, Sand.ai recorded 17.4 million yuan — about $2.61 million — in revenue in the first half of 2026, and a net loss of about 152 million yuan, or $22.8 million. The numbers underline how early the business still is: a billion-dollar valuation on a few million of half-year revenue.
## What Sand.ai Builds
Sand.ai was founded in early 2024 by Cao Yue, a Tsinghua PhD graduate who co-authored the Swin Transformer paper and previously co-founded Light Year, which Meituan acquired in June 2023 for $281 million. The startup builds foundation models for video generation, betting that video is the most direct path to a world model.
Its models so far are Magi-1, an autoregressive video generation model, and GAGA-1, for simultaneous audio-visual generation. The next step is MAGI-2, a mixture-of-experts system of roughly 114 billion parameters; a preview version is already open-sourced, and the official release is coming soon.
On the application side, its VidMuse music-video service passed $10 million in annual recurring revenue within two months of launching in January. That is a run-rate claim, not recognised revenue — but it is the closest thing Sand.ai has to proof that people will pay for what its models make.
## Our Read
The valuation is the story, and it is also the question. Sand.ai’s half-year revenue was about $2.6 million against a loss of $22.8 million, yet investors have priced it above a billion dollars. The bet is on MAGI-2 and the world-model thesis: whoever owns the best video generator owns the training ground for machines that understand the physical world.
The funding climate is helping. This has been a big week for startup money: [Zipline raised $1.5 billion for drone delivery](https://geroundtable.com/zipline-1-5-billion-drone-delivery-network/) and [Sequoia raised $10 billion in new funds](https://geroundtable.com/sequoia-10-billion-funds-london-shakeup/) while shrinking itself. Sand.ai is a smaller cheque, but it is the same signal — capital is still flowing to AI companies with a credible claim on the next platform.
The comparison that should worry Sand.ai is Kling. Kuaishou’s video unit closed a roughly $2.9 billion raise in October at a valuation near $18 billion, with second-quarter revenue above $116.9 million, up over 200% year on year. Sand.ai is chasing a rival with real revenue and eighteen times its valuation — with a model that has not officially shipped yet.
## Why It Matters
Video generation is where the AI race is heading next: the models that can render the world convincingly are also the models that can simulate it. A $200 million round for a two-year-old company with a preview model shows how seriously investors take that idea — and how little patience they have for waiting.
China has also adopted labelling rules for AI-generated synthetic content, with suspension or penalties for non-compliance. The market Sand.ai is entering comes with rules attached — and a model that can fake reality convincingly will be judged on how honestly it labels what it makes.
## Background
Sand.ai is the third funding story of its kind this week, but its model-first strategy sets it apart from the delivery and infrastructure rounds around it. The open-weights playbook is spreading: [Thinking Machines’ Inkling 2 Small is also arriving closed first, with weights to follow](https://geroundtable.com/thinking-machines-inkling-2-small/). For Sand.ai, the test is whether MAGI-2’s release justifies a valuation built mostly on promise.
Source: [TechInAsia’s report](https://www.techinasia.com/news/chinese-video-startup-sand-ai-raises-200m) and [DealStreetAsia’s report](https://www.dealstreetasia.com/stories/sand-ai-enters-unicorn-club-500283)
0 Comments