Alibaba to Charge Large Commercial Users of Next Open-Source AI Model

Revenue-Sharing Model Emerges

Alibaba plans to require major commercial users of its upcoming Qwen3.8-Max open-source AI model to share a portion of their revenue with the company, two people familiar with the strategy confirmed. The move follows a similar licensing structure introduced by Chinese AI startup Moonshot for its Kimi K3 model released in July.

The Qwen3.8-Max model is an open-weight system, meaning developers can download the underlying parameters and adapt the technology for their own applications. While Alibaba has historically charged for models hosted on its cloud platform, most of its open-source releases have been available without payment when deployed in customers' own data centers. That approach is changing.

The company intends to implement the new revenue-sharing requirement when it releases the model next week. The exact percentage Alibaba will claim remains under negotiation, though Moonshot has set a precedent by requesting up to 30 percent of revenue from partners whose annual sales from the model exceed 20 million dollars.

Following a Silicon Valley Playbook

The licensing terms reveal a maturing business model among Chinese AI developers, who are adopting tactics long used by US software companies: distribute technology at little or no initial cost, then charge for heavy commercial use and premium services.

Moonshot's Kimi K3 license includes a provision requiring any entity that offers the model as a commercial service and generates more than 20 million dollars in annual sales to negotiate a commercial agreement with the company. Chinese IT services provider Chinasoft International disclosed a revenue-sharing agreement with Moonshot in a July regulatory filing, though it did not specify the percentage.

Paddy Srinivasan, chief executive of cloud computing firm DigitalOcean Holdings, which offers Kimi K3 and other Chinese models to customers, explained the value proposition. His company and others pay AI labs for optimization support, early access to model updates, and collaboration on deployment strategies. The models themselves may be open-weight, but the expertise and ongoing development justify commercial arrangements.

Cost Advantage Over US Rivals

Chinese open-source models are competing aggressively on price. Kimi K3 costs roughly one-third as much as Anthropic's Fable model when comparing listed prices for input and output tokens, the basic units of AI computation. This pricing pressure comes even as US officials have accused Moonshot of stealing technology from Anthropic, a claim Chinese authorities have rejected.

Revenue-sharing deals between US firms and Chinese AI labs are taking shape despite geopolitical tensions. Several American cloud providers now offer Chinese models, creating commercial relationships that span the Pacific even as Washington tightens technology export controls.

Dan Fu, vice president of kernels at Together AI, said companies in his sector make money through optimization rather than model licensing fees. Better token usage, faster inference, and application-layer improvements create value that customers will pay for, regardless of whether the underlying model is free to download.

Diverging from US Closed-Source Approach

The Chinese strategy contrasts sharply with dominant US players. OpenAI, Anthropic, and Alphabet's Google maintain closed-source models, keeping their underlying architecture proprietary. Open-source releases from US firms have typically lagged in capability, though that gap may be narrowing.

Thinking Machines Lab, a US AI startup, released its first open-source model in July and is expected to follow with more powerful versions. Lin Qiao, CEO of Silicon Valley-based Fireworks AI, said she sees no fundamental barrier to powerful open-source models from American developers. The question is timing and strategic commitment.

Market Share Battle Intensifies

Alibaba's move to monetize open-source users reflects broader competitive dynamics. Chinese AI firms have released models nearly matching the performance of OpenAI and Anthropic systems, but at significantly lower price points. By keeping initial access free or cheap while claiming a share of commercial success, they can rapidly build market share without sacrificing long-term revenue potential.

The approach also hedges against uncertainty about which applications will prove most lucrative. Rather than betting on specific use cases, Chinese labs are positioning themselves to benefit from any commercial success their models enable, wherever it occurs.

For developers and enterprises, the trade-off is straightforward: access to capable, affordable models in exchange for sharing upside if deployment scales. The model works particularly well in Asia, where cloud infrastructure costs remain high and enterprises are eager for alternatives to US-dominated AI platforms.

As Alibaba prepares to roll out Qwen3.8-Max next week, the licensing terms will test whether this revenue-sharing framework can become standard across the industry. If successful, it may redefine how open-source AI is monetized, blending the accessibility of open code with the recurring revenue streams venture capital demands.