appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Vincent Chow

In China’s hyper-competitive and lucrative tech industry, releasing cutting-edge artificial intelligence models for free may seem counter-intuitive – but it has become a core business strategy.
At the University of Hong Kong last November, Alibaba Group Holding chairman Joe Tsai was asked why the tech giant open-sourced its AI models. Addressing a room full of students, Tsai said he believed open-source AI would bring global benefits by lowering costs, making it a natural fit for talent- and cash-poor countries.
The company’s flagship AI model, Qwen, has achieved nearly 1 billion cumulative downloads over the past three years, making it by far the most popular open-source model family worldwide.
Tsai’s interviewer Tang Heiwai, associate dean of the university’s business school, wanted to know what that meant in terms of profit. “How do you guys make money by being so generous?” he asked, smiling.
“We don’t make money from AI, that’s the answer,” Tsai responded, referring to the company’s models. His matter-of-fact remark reflects the company’s stance: while Alibaba, owner of the South China Morning Post, monetises inference and cloud services, it has still made open-source AI a central part of its corporate identity.
The question of how to monetise these models, which by their nature hold no intellectual property protections that would allow for straightforward profit-making, has been a recurrent one over the past year.
In China, the emergence of DeepSeek early last year sparked a wave of open-source models, with domestic start-ups MiniMax and Zhipu AI riding the wave all the way to blockbuster listings in Hong Kong.
The question has resurfaced in recent weeks during earnings season, as several companies – including Alibaba and Zhipu – released some of their latest models as closed-source software.


This followed the high-profile departure of Qwen’s technical lead, Lin Junyang, a fervent open-source advocate, prompting speculation that Alibaba was abandoning its previous strategy in favour of monetisation.
But open source and monetisation are not necessarily in conflict, several analysts said in interviews.
The key, said Daniel Yue – an assistant professor at the Georgia Institute of Technology’s Scheller College of Business, who researches open-source business strategies – is recognising that AI models are only one layer in a product’s broader “stack”.
Most AI companies, he noted, are pursuing a hybrid approach that combines open- and closed-source elements.
An old playbook
AI models are served to users through a process called inference, in which powerful graphics processing units (GPUs) provide the computational power to run them. While open-source models are made freely available, deploying and fine-tuning them on local hardware can cost users thousands of US dollars and demands a great deal of technical expertise.
Instead, model developers often provide the inference themselves – a form of “indirect monetisation”. In his November talk, Tsai compared this to staying at a hotel: rather than building it themselves, customers effectively rent one and pay for additional services, such as room service or spa treatment.
Similarly, Chinese cloud giants such as Alibaba and Tencent Holdings monetise open-source models by renting out their GPUs to users.
“Most people don’t want to deal with installing, updating and debugging an open-source version of anything,” said Kevin Xu, founder of technology hedge fund Interconnected Capital and a former senior director at open-source platform GitHub.
Chinese players are ... trying to make their models as abundant as possible, then monetising them in a more incremental way
Zhang Jiang, freelance consultant advising AI startups
Inference providers can also charge enterprises for premium features, such as security and optimisation. In 2002, legendary software engineer Joel Spolsky summarised this open-source business strategy as “commoditising your complement”: that is, making one layer of the value chain cheap or free to capture the value flowing to another layer.
The approach has succeeded with other foundational software, the most famous example being Google’s open-source Android for smartphones and other devices, which helped make it the default mobile ecosystem for billions of users. That, in turn, entrenched Google’s search engine, further cementing its grip on the market.
Alibaba’s cloud business offers another example. According to research firm Omdia, Alibaba Cloud holds about 36 per cent of China’s market share, making it the dominant player. Cloud revenue has been the company’s biggest growth driver since 2024, a year after it began open-sourcing Qwen.
But the strategy has limits. Zhang Jiang, a freelance consultant advising AI startups on capital raising and former director of equity research at UBS, said open-source strategies can be challenging, as intense competition – both domestically and from US models – puts downward pressure on prices.
According to Alibaba’s latest earnings report, its cloud computing margins remain in the single digits. By contrast, US companies that develop closed-source models, such as OpenAI and Anthropic, enjoy margins of about 40 to 50 per cent, as their exclusive control over inference gives them greater pricing power.
“What the US players are doing is a value-driven strategy of maximising every single dollar they spend on a per-user basis,” Zhang said. “The Chinese players, meanwhile, are pursuing a volume-driven strategy, where they are trying to make their models as abundant as possible, then monetising them in a more incremental way.”
In a market where the model is commoditised, price – and therefore cost efficiency – becomes the key differentiator, said Yue of the Georgia Institute of Technology. While third-party benchmark platforms like Artificial Analysis show these Chinese models still trailing US competitors by three to six months of development time, most of the country’s open-source models have no comparable gap among themselves, limiting the pricing power of any single player.
This leads to a familiar phenomenon, one which has already plagued China’s electric vehicle and solar panel industries: price wars. In their latest earnings, covering the year ending December 2025, both MiniMax and Zhipu reported widening losses despite triple-digit revenue growth. Research and development spending continued to balloon, while prices were kept low amid intense competition.




Still, both companies saw their share prices spike following their earnings, as attention turned to how a new phase of demand could reshape their business models.
A newer playbook
The emergence of the open-source AI agent tool OpenClaw at the beginning of the year has changed the calculus for Chinese AI companies, said Lu Shengyun, a Shanghai-based consultant who has advised both Zhipu and MiniMax on their commercialisation strategies.
OpenClaw autonomously executes tasks on behalf of users, who can choose which model to use with the agent. Once deployed, users can leave the tool running overnight to carry out tasks – an intensive process that could boost demand for AI models exponentially, Zhipu CEO Zhang Peng said at a recent industry event.
The development is positive for both cloud giants such as Alibaba and pure-play AI companies like MiniMax and Zhipu, Lu said. As demand rises, companies could gain more flexibility to raise prices and improve margins. It could also create the opportunity to shift from indirect to more direct monetisation, as higher demand reduces the need to offer free layers to build scale.
That has been the trend so far this year. While companies have not announced plans to abandon open-source strategies, executives emphasised a focus on model-as-a-service (MaaS) to investors during earnings season, a business model that charges users based on token consumption – the core metric of AI usage.
Alibaba CEO Eddie Wu Yongming has outlined plans to make MaaS its main AI business model, despite it currently contributing less than 10 per cent of its AI-related cloud revenue. For now, the company still relies on leasing GPUs to customers, but expects MaaS to become its largest source of cloud revenue within five years, with token consumption already rising sixfold in the first three months of the year.
Alibaba is not the only company targeting MaaS. Virtually every model developer in China has unveiled coding plans that allow programmers to pay fixed fees for heavy usage.
But amid a global GPU supply crunch, these limited-quantity plans have become almost impossible to buy, as companies restrict new sign-ups or, in Alibaba’s case, cancel the cheapest pricing tier less than two months after launch.


For Zhipu – one of the first Chinese companies to roll out a coding plan – the shift this year has been even more pronounced. In his earnings call, CEO Zhang Peng said the volume of requests through its application programming interface (API) had jumped 400 per cent in the first quarter, despite raising call prices by 83 per cent.
“Many vendors and customers who previously attempted to deploy open-source models locally have gradually or partially shifted to using our cloud-based model APIs,” he said. “This has made API and token billing our main business model now.”
Hybrid approaches
OpenClaw kick-started a shift in strategy that could put the industry on a more sustainable footing, potentially even giving some open-source model developers pricing power they previously lacked. But competition remains fierce, Yue said.
For smaller players such as Zhipu and MiniMax, continuing to open source their models poses a growing challenge: they are effectively handing over core tools to Big Tech competitors, as Alibaba, Tencent and other major cloud providers incorporate Zhipu and MiniMax’s models into their own coding services.
Early signs show how they might defend their market position. On Sunday, MiniMax finally open sourced its latest M2.7 model after a lengthy delay. But crucially, it amended its terms of use to prohibit commercial use of the model without authorisation.
The move triggered a backlash in the open-source community. Until recently, Chinese open-source models had largely followed open-source software practices, like Android’s, through licences that support enterprise adoption by granting commercial use without requiring compensation.
MiniMax’s updated licence has raised questions about whether their models can even be called “open source” any more. Florian Brand, a Germany-based analyst of China’s AI ecosystem, said licensing restrictions could undermine the popularity of Chinese models among the global developer community, given the wealth of alternatives available.
But moves to limit competitors by adjusting licences, Yue argued, are not new to the open-source software industry. “It’s not surprising that people are upset, but it’s also a natural way for companies to try to seek sustainability,” he added, noting that open-source companies must balance short-term reputational benefits with long-term commercial viability.
For Brand, the long-term viability of Chinese companies offering open-source AI models remains unclear. He pointed to the shortfall in computing power in China compared with the US, which could spur companies to save resources by prioritising leading closed-source models.
A major variable is DeepSeek’s next release, rumoured for this month. The Hangzhou start-up’s R1 model, released last year, kick-started the open-source wave by lowering prices and eliminating any pricing advantage its competitors had. Will its next model do the same again, forcing companies to return to a more open-source approach? “We just have to wait and see,” Brand said.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

China’s AI firms scaled up on open-source models. The next phase may be different

Illustration: Henry Wong

‘Common ignorance’: how China took the lead in global efforts to govern AI’s future

China is leading a global charge to rein in AI as governments and industry leaders fear the world is entering the unknown in terms of where the technology is going. Photo: Shutterstock

In a panel discussion on Saturday, Fu Hongyu, a policy lead at Alibaba Group Holding, said China is at the front lines of global efforts to introduce guardrails surrounding fast-changing AI technology. Photo: Handout

China’s Zhipu AI open-sources flagship model, raises prices to narrow gap with US rivals

The move marks Zhipu’s second price increase this year, following a broader overhaul in February that lifted rates by more than 30 per cent. Photo: Shutterstock

Zhipu’s GLM-5.1 is now priced closer to its US counterparts, though a big gap remains. Photo: Shutterstock

China’s edge over US in AI world models: abundant data, faster deployment, executive says

GigaAI, a company specialising in embodied intelligent models, has secured another round of financing. Photo: Handout

Founded in 2023, GigaAI is one of China’s earliest ventures to experiment with world models. Photo: Handout

Alibaba cloud growth forecast to accelerate with AI push, higher service charges: analysts

The Alibaba building in Xuhui Binjiang Park, also known as “AI Park”, in Shanghai. Photo: EPA

Price increases are expected to help boost long-term cloud margins of Alibaba, according to analysts. Photo: Reuters

China quietly profits from US AI boom despite Washington’s tech curbs: research

Roughly US$2 trillion worth of data-centre projects are planned or under way in the US. Photo: Getty Images

China’s acumen in areas such as production of printed circuit boards allows it to benefit indirectly from the AI boom, according to Oxford Economics. Photo: Reuters

Tech

Tech Trends

Big Tech

Business

Companies

News

China

Free models helped Chinese firms find rapid success – but thin margins are forcing a shift towards hybrid business strategies


Free models helped Chinese firms find rapid success – but thin margins are forcing a shift towards hybrid business strategies.


China’s AI firms scaled up on open-source models. The next phase may be different

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Free models helped Chinese firms find rapid success – but thin margins are forcing a shift towards hybrid business strategies

Free models helped Chinese firms find rapid success – but thin margins are forcing a shift towards hybrid business strategies