The “Free” Era Is Ending: China’s Open-Source Giants Are Quietly Attaching a Price Tag to the Weights

First DeepSeek's surge pricing. Now Alibaba and Moonshot want a cut of your revenue. Three Chinese labs, one window, one conclusion: the "give the weights away" strategy is being rebuilt into a royalty business — and watchchina.ai called it yesterday

The Pattern Becomes Undeniable

Yesterday, watchchina.ai reported DeepSeek's 355% price hike and its invention of surge pricing for AI, and argued it marked the first retreat in China's price war — the beginning of the monetisation reckoning this portal has forecast since the 140-trillion-token analysis. One data point is an event. Three data points in the same window is a pattern. And the pattern is now undeniable: China's open-source AI leaders are, together and almost simultaneously, attaching a price to what used to be free.

The evidence extends well beyond DeepSeek. Alibaba plans to require large commercial users of the open-weight version of its Qwen3.8-Max model to share a portion of the revenue they generate from it. And Alibaba is not the pioneer here — it is a follower. The move follows a similar licensing approach taken by Chinese AI startup Moonshot for its Kimi K3 model: Moonshot's terms require any party selling Kimi K3 as a service and generating more than $20 million in annual revenue to reach a commercial agreement with Moonshot, with reported ceilings reaching up to a 30% revenue share. Read the three moves together — DeepSeek's surge pricing, Moonshot's revenue threshold, Alibaba's revenue-share plan — and the conclusion writes itself. As one industry tracker put it precisely: two Chinese labs converging on the same freemium structure in the same window matters more than either move alone; it suggests the "give the weights away, monetize the inference" playbook is being replaced with something closer to a downstream royalty.

What "Open" Means Now — The Crucial Distinction

watchchina.ai has documented China's open-weight strategy as its single greatest weapon across dozens of briefings, so this shift demands a precise understanding of exactly what is — and isn't — changing. The models are not becoming closed. The weights of Qwen3.8-Max and Kimi K3 remain freely downloadable; a student, a researcher, or a small developer can still take them and run them at no cost. What is changing is the treatment of the largest commercial users — the companies building serious, revenue-generating businesses on top of these models. Those users now face a new reality: the model is free to download, but not free to build a fortune on. Above a revenue threshold, you must negotiate, and you must share.

This is the "freemium" model arriving in open-source AI, and it is a genuinely clever resolution to the contradiction watchchina.ai has flagged all summer. China's labs spent enormous sums making their models popular by giving them away — and then watched American cloud providers like DigitalOcean, Together AI, and Fireworks AI build commercial services on top of those free Chinese models, capturing the value while the Chinese labs captured only market share. The revenue-share model closes that leak. It lets China keep the diffusion advantage — the free weights that spread Qwen and Kimi across the planet and drove the ecosystem lock-in — while finally collecting from the downstream businesses that turned that diffusion into profit. Diffusion at the bottom, royalties at the top. It is the barbell watchchina.ai forecast, formalised into a licence.

The Deeper Meaning — and the Honest Complications

The strategic significance is that China's AI industry is maturing from a land-grab into a business, and the timing is not coincidental. Moonshot just restructured as a joint-stock company ahead of a Hong Kong IPO, which raises the pressure to show recurring commercial revenue — the same capital-markets discipline watchchina.ai has tracked pushing Chinese labs toward sustainability all summer. The free era was the customer-acquisition phase. The royalty era is the monetisation phase. This is exactly the sequence any dominant platform follows: capture the market with free, then convert the market to paid once switching costs are high enough that customers cannot easily leave. Having spent to make Qwen and Kimi indispensable, China's labs are now, rationally, moving to collect.

But watchchina.ai will not present this as a clean victory, because it carries a real and specific risk that cuts against China's own interests — and it opens a door for its rivals. The moment "open" Chinese models come with a variable revenue-share attached, they lose part of what made them irresistible: total, frictionless, cost-certain freedom. As one analysis noted bluntly, if you are building a commercial product on Qwen or Kimi, your gross-margin model now has a variable slice you do not own — and that opens a real lane for Meta's Llama and other permissively licensed alternatives to pitch enterprise buyers on licence certainty. Here is the irony watchchina.ai finds genuinely striking: for a year, the story was American models locking down while Chinese models stayed free. Now, at the very moment America's Nvidia is funding open-weight models to imitate China's diffusion playbook, China's own labs are attaching commercial strings that make their models feel a little more like the closed Western products they disrupted. The convergence this portal has tracked all summer is not one-directional. Both sides are drifting toward the same pragmatic middle — free enough to spread, paid enough to survive. The "free" era of Chinese AI is not ending because China failed. It is ending because China won the market it set out to win, and is now, at last, sending the invoice.

Sources: Reuters, TechNode, Quartz, Business Standard, The Next Web, AI Weekly, Artificial Intelligence News