Alibaba Releases Wan 3.0 and Transforms $10 Billion Into a New Bet on Corporate Video AI
One day after the largest secondary offering ever made in Hong Kong, the Chinese group launched a model that generates 30-second videos from PDFs and spreadsheets, priced at $0.05 per second in 480p.
Alibaba announced the release of Wan 3.0 on Monday, its new AI video generation model, less than 24 hours after closing a secondary issuance of $10.2 billion in Hong Kong. This marks the largest follow-on primary offering in the history of the Asian stock exchange, with every dollar raised earmarked for AI infrastructure, aimed at competing with ByteDance, OpenAI, and Google in the generative video sector.
The new model doubles the output duration compared to Wan 2.7, reaching 30 seconds in a single pass, with resolutions of 480p, 720p, and 1080p. The pricing table published on Qwen Cloud lists $0.05, $0.10, and $0.20 per second of generated video, respectively. The access quota for approved accounts allows for two simultaneous requests, 30 calls per minute, and an asynchronous queue of 50 tasks.
The difference that matters to corporate buyers is not in clip length, but in what the model accepts as input. Wan 3.0 can read .doc, .xls, .ppt, .pdf, .md documents and web pages, converting these files into video. A results presentation, a quarterly report, or a training slide deck now becomes direct raw material for an audiovisual production agent. This category, internally referred to by Alibaba as Video Production Agent, targets clear market segments: advertising agencies, corporate marketing, internal training, and the Chinese mini-drama industry, which was already producing with Wan 2.7.
The CapEx Account Explains the Urgency
The launch of Wan 3.0 comes during a challenging quarter for Alibaba. Net profit fell by 75%, down to 10.44 billion yuan, while CapEx surged 75%, reaching 67.68 billion yuan. Management chose to turn this gap into an opportunity: it issued stock rather than taking on debt, signaling to the market that the race for AI infrastructure is lengthy, and that the company will secure space in the schedules of suppliers like TSMC, SK Hynix, and Samsung before prices rise further.
This is the second time in just over two weeks that Alibaba has shaken up the generative video market. On August 6, Wan 3.0 entered public beta. On the 7th, the company released open weights, but stopped at Wan 2.2. The weights for 3.0 remain closed, available only via API on Qwen Cloud and in Model Studio. The strategy divides the portfolio between an open track for researchers and a proprietary track for direct monetization, mirroring the approaches taken by Meta and Mistral in the language category.
Where the Story Lands
For the Japanese market, the arrival of Wan 3.0 impacts advertising production costs. Major agencies like Dentsu and Hakuhodo had been experimenting with large-scale video production using Runway and Pika, whose cost per second in 1080p is several times what Alibaba now charges. At $0.20 per second in 1080p, it sets a new price benchmark for the entire Asian media supply chain.
In India, the effect is different. Advertising content studios in Mumbai and Bengaluru provide low-cost video services to American and European clients. The document-to-video capability of Wan 3.0 compresses the most expensive step of this process, the storyboard and initial breakdown. This pressures the margins of local suppliers and accelerates the narrative of the agentic media production divisions that WPP, Publicis, and Indian consultancies have begun to establish by 2026.
What the CIO Needs to Decide
The Chinese model may not surpass the technical benchmark set by Google's Veo 3 or OpenAI's Sora 3, but it resolves a problem that Western competitors still treat as a lab project: accepting corporate files as input. According to data released by Alibaba itself, the two-week beta was used in short drama productions, advertising campaigns, tourism promotions, and music videos, suggesting that the company is calibrating the model for commercial use cases before scientific refinement.
The unresolved question for global chief technology officers is one of dependency. Adopting Wan 3.0 means depositing corporate data, often containing sensitive marketing material, onto Alibaba Cloud servers subject to Chinese jurisdiction. Western companies exposed to the European AI Act will have to decide whether the savings of $0.15 per second warrant the compliance risk or if they should restrict usage to APAC environments for non-confidential content. Western competition gains some breathing room to react, but the price floor for the category has just been reset, and the segment of the supply chain that relies on storyboarding has lost protection.