Innovation culture is in our DNA, says Alibaba chairman

At the VivaTech conference in Paris this week, Alibaba chairman Joe Tsai spoke candidly about the challenges the company has faced in recent years, and where it’s headed in the future.

Alibaba has gone through several changes in recent years. During the Covid-19 pandemic, the company faced challenges including regulatory scrutiny, geopolitical tension and “stiff competition”, he said.

At the tail-end of pandemic, Tsai, whose big passion is basketball, wanted to phase himself out of the company and focus on family and sports instead. However, it wasn’t long before he returned to Alibaba as its chair in 2023. “I think I saw a company that kind of lost its direction a little bit,” said Tsai, explaining that the company had grown too big and needed to “shrink the size of our balance sheet”.

He added that while the geopolitical environment today is “still very difficult, very challenging”, the company has two clear focuses: e-commerce and cloud computing, “with a heavy artificial intelligence [AI] element injected into the cloud business”.

“We want to have AI everywhere,” said Tsai. “AI both within the company to help out efficiency, productivity wise, but also have AI infused into every app that we make.”

He also highlighted the creativity both at Alibaba, but also in the Chinese technology market as a whole. “When you look at the Chinese market, starting with some of the early internet companies, China has always been very good at iterating and adopting new technology, especially in the consumer internet space,” said Tsai.

“You go to China, all the consumer internet apps, it’s not YouTube, Google, Facebook or Instagram,” he said. “It’s the Chinese version of those, and in fact, they’re actually better. They provide a better user experience. So, I think China has a very application-rich ecosystem, and companies are very willing to adapt these applications.”

He added that there is also a creative and competitive culture. In January 2025, the world was stunned by Chinese startup DeepSeek, which caused waves with its highly economical large language model (LLM). The release of DeepSeek R1 caused a global sell-off of AI companies’ stock, and at Alibaba, the pressure was mounting.

As the DeepSeek model took off, it was close to Chinese New Year, where the country effectively shuts down for two weeks. However, at Alibaba, there was no time to lose.

“What happened was our engineering leads decided to cancel their Chinese New Year holiday, and told everybody to stay in the company, sleep in the office: ‘We’re going to accelerate our development’,” said Tsai.

“Within a few weeks, we developed and launched our own version, which is the Qwen series of LLMs.”

Qwen 2.5-Max is a mixture of experts (MoE) model pre-trained on 20 trillion tokens, and post-trained with curated supervised fine-tuning and reinforcement learning from human feedback.

MoE is a technique in which a model is structured with multiple “minds”, and each mind is compartmentalised so that whenever there is a query, the model uses adaptive routing to go to the specific mind, or region, that has the answer. For example, if a model is geared towards coding, the model routes queries to that mind.

“It’s not bad, it’s very competitive,” said Tsai, adding that the company decided to open source it, explaining that the company believes it makes it easier to build up an ecosystem where it will drive the demand for AI and drive training needs to make it better.