Alibaba-backed video push
Alibaba’s cloud arm led a $293 million funding round for ShengShu and a stealth Alibaba video‑generation model reportedly debuted atop global benchmarks, signaling faster, better competition in AI video technology. The funding and the benchmark claims suggest non‑U.S. players are moving aggressively to own higher‑quality video generation capabilities that could reshape supplier choices for newsroom tooling. (x.com)
Alibaba just put real money behind AI video twice in the same week: its cloud unit led a 2 billion yuan funding round for ShengShu Technology, and Bloomberg reported Alibaba also built a separate video model called Happy Horse that landed at the top of Artificial Analysis’s text-to-video leaderboard on debut. (bloomberg.com 1) (bloomberg.com 2) The funding round was about $293 million at current exchange rates, and Bloomberg said Baidu Ventures and Luminous Ventures joined Alibaba Cloud in backing ShengShu. ShengShu said it plans to use the cash to build what it calls a “general world model,” which is software meant to simulate how people, objects, and camera motion behave across a scene. (bloomberg.com) ShengShu is not a random lab. Its best-known product is PixVerse, an AI video app that Bloomberg reported raised another $300 million in March 2026, which means Alibaba now has exposure to both an outside video startup and its own in-house video research. (bloomberg.com) The benchmark claim matters because AI video is now judged less like a demo reel and more like a sports table. Artificial Analysis runs a public leaderboard based on blind preference votes, and its page shows HappyHorse-1.0 at number one with an Elo score of 1,355, ahead of ByteDance’s Dreamina Seedance 2.0 at 1,273 and ahead of OpenAI’s Sora 2 Pro at 1,196. (artificialanalysis.ai) Alibaba already had a video line before Happy Horse appeared. Alibaba Cloud’s Wan models were opened up in February 2025, and the company said then that Wan2.1 was the only open-source video generation model in the top five of the VBench leaderboard, which is another benchmark used to score motion, prompt following, and visual quality. (alibabacloud.com) (github.com) By December 2025, Alibaba had pushed that line to Wan 2.6, adding tools that let users generate multi-shot videos, keep a person’s appearance consistent, and clone their voice for dialogue scenes. Alibaba Cloud said those models were aimed at “professional-grade content production,” which is closer to ad-making and studio workflows than to novelty clips. (alibabacloud.com) Happy Horse looks like a different bet from Wan. Bloomberg reported it came out of Alibaba Token Hub, is still in beta, and is expected to get application programming interface access soon, which usually means Alibaba wants outside developers and software vendors to build on top of it instead of keeping it as an internal showcase. (bloomberg.com) That changes the map for buyers because the strongest video tools are no longer clustered in the United States. Artificial Analysis’s current top ranks are crowded with Chinese names including Happy Horse, Dreamina Seedance, SkyReels, Kling, PixVerse, and Vidu, while Google, OpenAI, and Luma sit further down the same table. (artificialanalysis.ai) Alibaba is also spending at the infrastructure layer underneath all of this. In February 2025, Alibaba Group said it would invest at least 380 billion yuan, about $53 billion, in artificial intelligence and cloud infrastructure over three years, which gives its model teams more chips, more training capacity, and a built-in route to sell the finished systems through Alibaba Cloud. (alibabagroup.com) For newsrooms and media vendors, the immediate shift is not that one model won one chart. The shift is that Alibaba now has three pieces at once — cloud distribution, its own benchmark-leading model, and a large stake in a fast-moving startup — and that is the kind of stack companies use when they want to become the default supplier instead of just another lab. (alibabacloud.com) (bloomberg.com 1) (bloomberg.com 2)