Skip to content
NewsFunding & M&A

Alibaba's $10.2 billion Hong Kong share sale funds AI buildout as Wan3.0 video model ships

· by Pondero Newsdesk

The short version

Alibaba priced 710 million new HKEX shares at HK$112.70 each on August 23, raising HK$80 billion for its AI full-stack strategy, then launched Wan3.0 the following day: a video model that generates clips up to 30 seconds at 1080p from text, PDFs, and slide decks.

Alibaba's $10.2 billion Hong Kong share sale funds AI buildout as Wan3.0 video model ships

Alibaba priced 710 million new Hong Kong-listed shares at HK$112.70 each on August 23, closing Hong Kong's largest-ever primary follow-on offering and raising HK$80 billion ($10.2 billion USD) earmarked entirely for AI. One day later, Alibaba's Tongyi Lab formally launched Wan3.0: a video generation model that produces clips up to 30 seconds in a single pass and accepts PDFs, slide decks, and spreadsheets as creative inputs alongside text and images.

What happened with the share sale

The offering was priced at a 3.6% discount to the August 21 close, per Taipei Times citing company filings. Settlement is scheduled for August 26. Institutional investors requested nearly three times the shares on offer, per Bloomberg.

The placement ranks as the largest primary follow-on offering in Hong Kong Stock Exchange history and the third-largest globally in 2026, behind follow-on offerings from Alphabet and Intel, per Bloomberg. Alibaba directed 100% of proceeds to what it called its AI "full-stack" strategy, covering chip procurement, infrastructure expansion, and model development.

Chairman Joseph Tsai purchased approximately HK$80 million of shares in the offering; CEO Eddie Wu purchased approximately HK$40 million, per Taipei Times. The placements came alongside Alibaba's standing three-year commitment to spend 380 billion yuan ($56.5 billion) on AI infrastructure, chips, and data centers. Alibaba stock fell 8.5% on August 24 in its largest single-day decline since early 2025, per CNBC.

What Wan3.0 ships

Alibaba's Tongyi Lab formally launched Wan3.0 on August 24. A public beta had been running since August 6, giving the model roughly three weeks of production exposure before its general announcement.

The central capability upgrade over Wan2.7 is single-pass video generation up to 30 seconds, double the predecessor's 15-second ceiling. Output resolutions run 480p, 720p, and 1080p. Alibaba Cloud's Model Studio prices the API at $0.05 per output second at 480p, $0.10 per second at 720p, and $0.20 per second at 1080p, per The Next Web. At 1080p, a full 30-second clip costs $6 before any discount. From August 24 through September 23, selected platforms carry a 30% temporary discount on Wan3.0 API calls, per The Next Web.

The input surface is broader than Wan2.7. Beyond text and images, Wan3.0 accepts audio, existing video clips, web pages, PDFs, spreadsheets, and PowerPoint files as creative references. Alibaba described the document-to-video capability as enabling conversion of "static text-heavy data" into video, per The Next Web, with production use cited in short drama, advertising, tourism promotion, and music video work during the beta period.

One notable change from earlier Wan releases: Wan3.0 ships with no open weights. Previous versions (Wan2.1, Wan2.7) were available on Hugging Face with downloadable checkpoints and ComfyUI support. Wan3.0 is API-only, with access gated through an application process on Model Studio and Qwen Cloud, per The Next Web. No Hugging Face repository or public checkpoint exists as of launch.

Why it matters

For teams running video generation workflows, Wan3.0's 30-second single-pass ceiling matters more than it might appear. Most competing video APIs still cap single-pass generation at 10 to 15 seconds, requiring multi-shot stitching for anything longer. Wan3.0's pricing at $0.20 per second at 1080p puts a 30-second clip at $6, a price point that lands it as a viable option for advertising formats where multiple takes are standard.

The document-to-video capability is the genuinely new angle. Accepting PDFs and slide decks as source material is not a default feature of RunwayML, Kling, or Google Veo. If the pipeline is reliable in practice, it offers marketing and editorial teams a direct path from existing content assets to video without a separate scripting step. That changes the staffing math for teams that currently need a producer to reformat research decks or reports for video distribution.

The closure of model weights is a strategic shift worth tracking. Open Wan releases built developer communities on ComfyUI and drove adoption at a scale that closed models typically cannot replicate quickly. Alibaba is trading that community flywheel for direct API revenue control. Whether that trade is correct depends on how fast enterprise contract volume grows relative to the organic adoption the open weights would have generated.

On the capital side, the HK$80 billion raise is the largest AI-earmarked single-event capital deployment by a HKEX-listed company on record. The oversubscribed institutional book and personal executive participation signal conviction inside the company. The 8.5% stock decline on August 24 reflects market concern about dilution, not necessarily a rejection of the AI strategy.

Context

Alibaba's Qwen model family is reported as the most-downloaded globally, per Bloomberg. Alibaba accelerated its projected payback period on AI capital spending to 2.5 years from 3 years earlier in 2026, citing growing demand for cloud AI services, per Yahoo Finance reporting.

In the video generation market, Wan3.0 competes directly against RunwayML Gen-3 Alpha, Kling, Google Veo, and ByteDance's Seedance. ByteDance has been the most aggressive competitor in the 30-second tier. Whether the Wan3.0 pricing or its document-input capability prompts a response from ByteDance is the clearest near-term market test of how seriously the competition views this launch.

What to watch next

Secondary market performance of the new HKEX shares in the weeks after August 26 settlement will indicate how confidently institutional investors outside Asia price Alibaba's AI thesis. On the product side, whether Wan3.0's document-to-video pipeline gains traction with B2B buyers is the key question; consumer adoption of video generation is well understood, but enterprise willingness to pay for document conversion at these rates is not. A ByteDance response on Seedance pricing would be the most concrete signal that Wan3.0 landed as real competitive pressure.

Sources