Alibaba Group’s latest AI video model has moved to the top of a global benchmark. The result highlights the growing competitiveness of Chinese firms in tools used for advertising, content creation, and entertainment.
Summary
- Alibaba Group’s HappyHorse-1.0 topped global AI video benchmarks shortly after its anonymous debut on Artificial Analysis.
- The model was later confirmed as part of Alibaba’s ATH AI Innovation Unit, driving gains in its Hong Kong-listed shares.
- Rival setbacks from OpenAI and ByteDance have opened room for Alibaba to strengthen its position in video generation.
The model, called HappyHorse-1.0, first appeared on the benchmarking platform Artificial Analysis on April 7 without disclosing its origins at the time. It quickly rose to the top of blind-test rankings across both text-to-video and image-to-video categories, drawing widespread attention across the industry.
Developers for the model later disclosed that it is part of Alibaba’s ATH AI Innovation Unit and is still under active development. The announcement was made through a newly created account on X, confirming the company’s involvement after days of uncertainty.
Before the disclosure, speculation around the identity of the developer had intensified, with guesses ranging from Tencent to Alibaba or even an independent team.
Following the confirmation, Alibaba’s shares in the Hong Kong market closed 2.12% higher on Friday shortly after the company confirmed its involvement in the project. The stock had previously gained 6.75% earlier in the week, led by a recovery in tech stocks as U.S.–Iran tensions showed signs of easing and speculation around the model intensified.
The company has been expanding its AI presence in China’s competitive market, including building on its Qwen large language model and chatbot ecosystem. It had also introduced video generation features earlier, though none achieved the same level of attention or ranking performance in such a short duration.
HappyHorse-1.0 can improve Alibaba’s standing in video generation, a segment where rivals have faced hurdles recently. One of its competitors, OpenAI, recently discontinued its Sora video platform as it shifted toward coding tools, enterprise services, and AGI development due to high compute demands. Meanwhile, ByteDance paused the rollout of its Seedance 2.0 model following copyright disputes with major studios and streaming platforms.
Chief Executive Officer Eddie Wu has placed AI at the centre of Alibaba’s long-term strategy, which includes investments in chips, cloud infrastructure, and data centres. The company has a track record of embedding its models into e-commerce, advertising, and entertainment services, and a similar integration path could be considered for HappyHorse.
Alongside developments in video generation, Alibaba is expanding its underlying infrastructure. The company is working with China Telecom on a new data centre project in southern China, part of a national effort to strengthen domestic computing capabilities.
The facility will deploy 10,000 of Alibaba’s Zhenwu AI chips, designed to handle both training and inference workloads. The system is built to support models with hundreds of billions of parameters, placing it among the most advanced clusters currently in operation.
Alibaba said the chips can function as a unified system, allowing the cluster to operate like a single supercomputer with a latency of about four microseconds, improving efficiency for large-scale AI tasks.
In a separate development, Alibaba Cloud disclosed it had led a funding round of roughly $275 million for Chinese AI startup Shengshu Technology. The round also drew participation from Baidu Ventures and Luminous Ventures, following an earlier RMB 600 million raise completed just two months ago.
Shengshu Technology develops video generation tools through its Vidu platform, competing with products from ByteDance, Alibaba, Kuaishou, and emerging startups such as PixVerse. The investment signals continued competition in the segment, even as infrastructure and model development accelerate in parallel.