Alibaba’s AI is a hit, but it’s hard to monetize it

Last year, titans of China’s tech industry received a rare invitation to meet Xi Jinping, the country’s leader, at the ceremonial Great Hall of the People.

But as the entrepreneurs — including the heads of national stars like telecom giant Huawei and growing companies like robotics start-up Unitree — gathered in neat rows before Mr. Xi, there was an unexpected figure among them: Jack Ma.

Mr. Ma, who founded e-commerce empire Alibaba in a cramped apartment in Hangzhou and became a rock star of China’s Internet era, has all but disappeared from the public eye after Beijing torpedoed a planned $34 billion public offering of one of his companies after criticizing regulators in 2020.

But now Alibaba has re-emerged as an AI powerhouse, creating one of the world’s most widespread AI systems.

Alibaba has made its most popular AI models open source and allowed others to freely use and modify them. This made its technology much cheaper than the proprietary systems of American competitors such as Anthropic and OpenAI, which helped the company attract users from around the world.

But it also raised a difficult question – how to turn this global popularity into a profitable business. It’s a challenge many open-source technology companies face, and one that divides Alibaba’s AI team.

Alibaba has become one of China’s most valuable companies thanks to its global e-commerce business. In recent years, domestic rivals with lower costs have outmaneuvered it, and it has had difficulty expanding overseas. After Mr. Ma faded from the public eye, a new generation of executives took over, adopting a lower profile while emphasizing Alibaba’s role within Beijing’s broader technology policies.

“Alibaba is well aware of its position as part of this team of national champions leading China’s assault on artificial intelligence,” said Kyle Chan, a fellow at the Brookings Institution. “But they also have a huge responsibility not to repeat what they did before.”

In retrospect, Alibaba gradually laid the groundwork to become an AI heavyweight. As its online shopping and logistics business expanded, the company built data centers capable of processing massive amounts of customer data around the world.

It also borrowed a page from Amazon, parlaying its e-commerce success and technology infrastructure into big cloud computing, giving Alibaba two essential ingredients for building AI systems: data and computing power.

“Alibaba was ahead of its time in many ways,” Mr. Chan said. “In many of these areas that are hot topics now, they were planted early.”

Alibaba introduced its family of AI models, Qwen, in 2023 and quickly made them open source. When DeepSeek, an artificial intelligence start-up, burst into the limelight claiming to have built a powerful model at a fraction of the cost of Western competitors, global investors jumped on the hype surrounding China’s open-source artificial intelligence.

China has made open source technology a pillar of its efforts to become an AI superpower. Almost all of its leading AI systems are open source.

In January, Qwen became the world’s most downloaded open-source AI system. Its models were downloaded about a million times a day, according to Hugging Face, a company that hosts many open-source artificial intelligence projects.

But that reach hasn’t translated into big bucks. In the first three months of this year, Alibaba reported $1.3 billion in sales from AI-related products — less than 4 percent of its total revenue. That pales in comparison to the company’s plan to spend more than $55 billion building its AI infrastructure by the end of next year.

The quest to turn its open source success into a profitable business has quietly fractured the team behind Qwen.

In March, Lin Junyang, Qwen’s chief engineer, announced that he was leaving Alibaba. Several other key engineers left around the same time. Two people familiar with the team said it split over disagreements over how best to commercialize Qwen.

Alibaba has long kept its largest and most advanced models proprietary while releasing leading open-source models alongside them. Now the company is signaling a broader shift away from widely used open-source models and toward closed ones that customers must pay to use. In April alone, Alibaba released three proprietary models within days of each other.

The Qwen team remains focused on keeping up with the leading models coming out of Silicon Valley, but there is a growing recognition that technology leadership will not be enough if the company does not make money, said a member of the Alibaba AI Research Lab, which includes the Qwen team, and another, who spoke on condition of anonymity because the person was not authorized to discuss internal matters.

The challenge for Alibaba is evident in its share price. While AI-related stocks have surged in markets around the world, Alibaba shares have fallen 37 percent this year in Hong Kong, where the broader market has fallen 12 percent.

Building a high-performance AI model is expensive and requires huge investments in hangar-sized data centers filled with computer chips that consume large amounts of electricity.

Chinese AI companies face another hurdle: US export controls limit their access to the most advanced chips. China’s top AI start-ups and researchers routinely say that the lack of computing power is the biggest thing holding them back, and they spend heavily to make up for it.

These exorbitant costs are piling financial pressure on Chinese AI companies. Start-ups such as MiniMax and Z.ai have gone public to raise money from investors, and Alibaba is increasingly steering customers toward proprietary models, even as the strategy erodes some of its best talent.

“It’s a tough choice that every open lab has to make at some point,” said Kevin Xu, founder of Interconnected Capital, a hedge fund that invests in artificial intelligence technologies.

Alibaba’s AI business is facing increasing external challenges. The Pentagon recently placed the company on a blacklist of firms it says support the Chinese military, a designation that Alibaba says has already hurt its business. Alibaba said in a statement that it is not affiliated with the Chinese military.

Anthropic and OpenAI have also accused Chinese companies, including Alibaba, of improperly harvesting data from their AI systems to speed up the development of their models.

Last month, Antropic sent a letter to Senators Tim Scott, R-South Carolina, and Elizabeth Warren, D-Massachusetts, accusing Alibaba of a “brazen” and “illegal” attempt to copy its technology using 24,000 fraudulent accounts. Alibaba declined to comment on the allegations.

Many in the tech industry compare today’s AI boom to the early days of the dot-com era, when Internet companies flourished but no one had yet figured out how to build lasting businesses on the technology.

Richard Lin, vice president of Silicon Valley-based Datastrato, who has long been involved in China’s open-source community, said all of China’s top AI companies are fighting for survival.

“Right now, there is no AI company with a sustainable business model,” he said. “It’s not a healthy industry.”