Sand.ai Raises $200M in Series A+, Reaches Unicorn Status

Beijing-based video foundation model startup Sand.ai has secured $200 million in a Series A+ round, cementing its unicorn status and accelerating development of MAGI-2, a 114-billion-parameter mixture-of-experts model that the company says generates 10-second 1080p video at roughly one-tenth the industry's prevailing per-second cost.

The round closed in early November 2026, bringing Sand.ai's cumulative fundraising to $300 million (approximately RMB 2.001 billion) over ten months — a pace that places it among the fastest-capitalizing pure-play video AI companies in China. The company declined to disclose its post-money valuation, but confirmed it has crossed the $1 billion threshold to become a unicorn. The timing is deliberate: Sand.ai says the full commercial release of MAGI-2 is imminent, and investors are effectively pricing in what comes next.

Initial market reaction underscores the competitive anxiety surrounding China's video AI race. With ByteDance's Seedance dominating via captive infrastructure and Kuaishou's Kling video model recently commanding a post-investment valuation of $18 billion (approximately RMB 130 billion) following a RMB 19.8 billion (US$2.75 billion) capital injection, Sand.ai's latest round signals that institutional capital is actively constructing an alternative bet — one with open equity and an open-source moat.

Cornerstone Capital and ChinaCache Lead a Strategically Layered Cap Table

The syndicate structure of this round is as analytically revealing as the dollar figure itself. Cornerstone Capital and ChinaCache International Holdings co-led the investment, joined by Times Capital, Panlin Capital, Dongzheng Capital, Tianchuang Capital alongside Shandong Railway Investment and Shandong Railway Fund, Pudong Ventures and Shanghai Guofang, plus returning investors Matrix Partners China, Huaye Tiancheng, and SICC.

Four distinct capital categories sitting on a single cap table is an unusual configuration. Cornerstone, historically conservative on early-stage large model bets, signals institutional conviction that Sand.ai has cleared the proof-of-concept threshold and now represents infrastructure-grade investment. ChinaCache's participation warrants separate scrutiny: the 17-year-old CDN provider disclosed in a late-September 2026 exchange filing that it committed RMB 300 million (US$41.7 million) for 2.1054 million Series A+ preferred shares. This is not passive financial exposure — the company explicitly cited co-development opportunities in audio-video inference, low-latency edge inference, and enterprise intelligent solutions. ChinaCache is buying distribution leverage into the next content delivery paradigm, not simply chasing AI valuation uplift.

Shanghai state capital — Pudong Ventures and Shanghai Guofang — adds a third dimension: regional industrial policy alignment. Beijing-headquartered Sand.ai attracting Shanghai government-backed capital reflects the cross-regional competition for AI cluster formation, a dynamic accelerating across Chinese tier-one cities in 2026.

Returning investors doubling down complete the picture. When early backers Matrix Partners China, Huaye Tiancheng, and SICC re-enter at a significantly higher implied valuation, they are issuing a mark-to-market endorsement that the original investment thesis remains intact — and that the ceiling has not yet been reached.

MAGI-2 Preview Validates Scaling Economics, Full Release Approaches

Sand.ai's technical trajectory traces a deliberate contrarian arc. Founded in January 2024 in Beijing's Zhongguancun district, the company launched when diffusion-based architectures dominated video generation orthodoxy. CEO Cao Yue — a 2019 Tsinghua University software engineering PhD, former principal researcher at Microsoft Research Asia, former visual model research center head at Beijing Academy of Artificial Intelligence, and co-founder of AI large model company Lightyear Beyond, which was subsequently acquired by Meituan — bet early on autoregressive generation as the structurally correct modeling approach for video data.

That bet has since compounded through three architectural iterations:

• April 2025: Sand.ai released and open-sourced MAGI-1, an autoregressive video generation model supporting continuous and long-horizon video generation via sequential temporal chunk processing.
• March 2026: The company open-sourced MagiHuman, a unified single-stream Transformer architecture for joint text, visual, and audio modeling, alongside infrastructure tools MagiAttention and MagiCompiler targeting large-scale training and inference efficiency.
• August 2026: Sand.ai released and open-sourced MAGI-2 Preview — the world's first video generation model scaled to 100-billion-plus parameters with openly released weights. Total parameters reach approximately 114B; single forward-pass activation engages approximately 6B parameters via an ultra-fine-grained MoE design. At launch, MAGI-2 Preview ranked sixth globally on Artificial Analysis's image-to-video leaderboard.

The inference economics embedded in MAGI-2 Preview deserve investor attention. Sand.ai reports a post-distillation cost of approximately RMB 0.5 (US$0.07) per 10-second 1080p video clip — roughly one-tenth the per-second cost of leading industry models. At that price point, the company is not merely competing on quality; it is actively compressing the switching cost for content platforms evaluating supplier diversification.

The forthcoming MAGI-2 full release will deploy at greater parameter scale than the preview version. The key question the market will use it to answer: can a Chinese startup close the remaining quality gap with global frontier models while maintaining its cost structural advantage?

VidMuse Hits $10M ARR, But the Revenue Conversion Test Remains Open

Beyond model benchmarks, Sand.ai has constructed an application-layer proof point. VidMuse, the company's AI video creation agent targeting content creators, crossed $10 million in annualized recurring revenue (ARR) within two months of launch in early 2026. For a foundation model company, this matters beyond the revenue line itself — consumer-facing products generate real-world usage data that feeds back into model training, creating the data flywheel dynamic that pure API businesses cannot replicate.

Cao Yue has stated explicitly that building only an API-sales model is insufficient in video generation, where model capability and product experience must co-evolve. This dual-track strategy — open-source foundation model for developer ecosystem building, proprietary application for direct monetization — mirrors the architecture that has proven durable in large language model competition.

However, $10 million ARR two months post-launch, while encouraging, does not yet constitute evidence of sustainable scale. The more demanding test arrives when VidMuse competes for professional content budgets against incumbents with deeper distribution networks. Sand.ai's ability to convert MAGI-2's cost and quality improvements into accelerating ARR growth will be the primary commercial metric watched by investors over the next 12 months.

Structural Market Logic: Why Independent Model Providers Gain Leverage as Giants Vertically Integrate

The competitive landscape is bifurcating in a manner that creates specific structural opportunity for Sand.ai. As leading platforms — including ByteDance and Kuaishou — integrate video generation capabilities vertically into content ecosystems, enterprise and professional content buyers face a concentration risk calculation. An advertising agency, film production house, or e-commerce content team generating video through a supplier that also competes in content distribution must weigh the risk of platform rule changes, proprietary data exposure, and dependency on a single vertically integrated vendor.

This structural tension is not hypothetical — it mirrors the dynamic that drove enterprise cloud customers toward multi-cloud procurement strategies and elevated the commercial value of independent infrastructure providers. Sand.ai, as a foundation model company without a competing content platform, can credibly position as a neutral supplier to any distribution ecosystem.

Kling's $18 billion valuation establishes the category ceiling that investors are triangulating against. The quantum gap between Kling's current valuation and Sand.ai's unicorn entry point simultaneously represents the magnitude of the prize and the distance yet to be covered.

Investment Implications: A Staged Bet on China's Video AI Infrastructure Layer

Sand.ai's $300 million ten-month fundraising run encodes a specific institutional thesis: that video generation is following the LLM commercialization playbook, that the category will ultimately support multiple durable winners, and that early positioning in a technically credible independent provider generates asymmetric returns regardless of whether the company finishes first.

The roster of prior-round backers — Look Capital, Lollapalooza Capital (the family office of Wang Huiwen, founder of Meituan's predecessor), Nine Chapters Capital, MSA Capital, Sinovation Ventures, Source Code Capital, IDG Capital, Baidu Ventures, and others — covers nearly the full spectrum of China's top-tier technology investment community. That spectrum does not come together around speculative theses.

The investment thesis resolves into three sequenced catalysts: MAGI-2 full release driving open-source developer community expansion and API commercialization; VidMuse ARR acceleration as model cost declines pull creation economics toward a mass-market inflection; and an eventual valuation re-rating as Sand.ai's scaling path becomes legible to later-stage and crossover investors.

What remains unresolved is execution risk. The company is 22 months old, operating in a segment where compute costs are substantial, iteration cycles are compressed, and ByteDance's Seedance operates with structural advantages in traffic, data, and capital that no independent startup can fully replicate. Sand.ai's founders have navigated that asymmetry thus far by choosing technical differentiation over head-on competition. Whether that strategy holds as MAGI-2 enters its full commercial phase is the defining question for 2027.

Related Coverage:

Seedance 2.0 Powers Volcano Engine's 10x MaaS Revenue Ambition

AI-Powered Games Flood China's Market, But Quality Gap Widens as Platforms Race for Scale