Tencent’s cloud boss plays long game to patch ‘leaky’ AI ship

When Pony Ma called Tencent Holdings’ artificial intelligence endeavor “a leaky ship” this summer, the billionaire founder sent ripples through a company that for years dominated the industry with WeChat. The social media and gaming leader was at the time struggling to catch the likes of DeepSeek in the post-ChatGPT age. One of his key lieutenants, Dowson Tong, came up with a surprising answer: Let’s retool to focus on businesses.

The Silicon Valley veteran focused the turnaround effort on something called WorkBuddy, then considered a side project within Tencent’s sprawling internet portfolio. By the fall, the Claude Cowork-style AI assistant, which combined the company’s existing messaging and office tools with AI agents that can take data and crank out reports, served 32 million monthly users — more than its rivals combined. The platform launched Tencent into a lucrative segment of AI, an important first step toward regaining its tech credibility.

Tong is now one of just a handful of key Tencent executives spearheading a high-stakes push to close the gap with TikTok-parent ByteDance and longstanding rival Alibaba Group Holding. At stake is Tencent’s future industry standing against a crop of fast-rising AI-native firms from DeepSeek to Moonshot AI, which’ve propelled the country to the forefront of the global AI race. WorkBuddy’s ascent signals a shift by Tencent to focus on enterprise agents rather than consumer chatbots, at least in the early phases, so it can better monetize an emerging technology, the Hong Kong native said in Shenzhen this week.

“It was an honest assessment,” Tong, 53, said of his boss’s frank comment. “Tencent’s advantage lies in our ability to correct course. If our first attempt doesn’t succeed and we recognize this as a marathon, we make a second attempt.”

Two decades after arriving in Shenzhen — barely able to speak Mandarin — Tong is helping guide Tencent in that competition against some of the world’s most innovative AI firms.

In a country where most people are used to low-cost, often free digital services, China’s AI pioneers struggle to translate billions of dollars of investment into sustainable profits. Yet the pressure from investors is growing — Tencent has shed 36% of its value this year to around $450 billion. That’s why Tong, who worked at Oracle early in his career, lit on deeper-pocketed corporate customers seeking reliability.

The turning point came when autonomous agent architectures such as OpenClaw began to gain traction with Chinese users at the start of this year. Tencent quickly launched its own variants. WorkBuddy — developed from an internal coding tool — captured a nascent enterprise AI market before ByteDance and Alibaba could respond. Tong drove that internal decision to entrust small, nimble teams to experiment with products — the same bottom-up playbook that produced some of the company’s biggest social media and gaming hits over its two-decade history.

“At the very beginning, a handful of people simply streamlined the interface and optimized it for general office work rather than developer tasks,” he said. “But when we saw strong word-of-mouth and capacity requirements climbing, we steadily ramped up our investment.”

Tong has built a career on similar comeback bets. After spending his early career in Silicon Valley, he was recruited in 2005 by Tencent President Martin Lau, a fellow University of Michigan alumnus. He started on Tencent’s earliest product, QQ, which he helped expand into blogging and mobile services.

Tencent CEO Pony Ma attends an event marking the 40th anniversary of China's reform and opening up at the Great Hall of the People in Beijing in 2018. | REUTERS

In 2018, he took the helm of a newly formed business group to offer Amazon Web Services-style cloud platform, lagging years behind Huawei Technologies and Alibaba. The company’s dominance in video streaming and gaming helped it court initial clients for its cloud business, and Tong also bet on video conferencing, transforming an internal tool into Tencent Meeting, whose popularity surged during COVID-19 pandemic lockdowns.

All that investment is starting to stretch Tencent’s finances. While its AI spending is nowhere close to the level of major U.S. tech firms such as Meta Platforms, the company’s free cash flow turned negative in the quarter ended in June. Tencent hiked its capital expenditure by about 180% during the period to about $8 billion.

Tencent remains intent on integrating AI into its primary businesses — Meta’s successful launch of Muse across Instagram and Facebook showed the industry the potential for artificial intelligence to reach an existing user base. The Chinese firm today counts more than a billion people who turn to WeChat to communicate, do business with merchants and play games.

Tong however noted that early efforts haven’t exactly stuck. In 2025, Ma personally ordered the integration of the company’s flagship chatbot — Yuanbao — with DeepSeek’s open-weight models, in the hope that the AI upstart could supplement Tencent’s lagging proprietary Hunyuan system. Yuanbao was subsequently embedded directly into WeChat.

But Yuanbao failed to deliver. Instead, ByteDance’s Doubao cemented its lead as China’s most popular ChatGPT-style app, surpassing 180 million daily users and converting an initial cohort into paying subscribers.

“I’m sure we’ll figure out a way to make the economics work,” he said. A key open question is when inference costs — the expense of hosting AI services — will fall enough for consumer products like Yuanbao to sustain profitability, Tong acknowledged.

Along with WeChat’s native agent, which is still in testing, Yuanbao remains a primary testbed for the company to develop an AI application that will ultimately rival TikTok or Tencent’s own super-app. Tencent is also open to exploring new formats, with Tong pointing to Meta’s Muse personal agent as a viable blueprint.

Now when the company’s top executives convene every two weeks, roughly a dozen of them conversing together, AI commands more than half of the agenda, Tong said. He also maintains close contact with Yao Shunyu, who joined from OpenAI to revamp Hunyuan’s development. Tencent’s next-generation Hy 4 model is slated to debut by year-end, setting the stage for another big trial.

The company isn’t slow, but simply bracing for a long fight, Tong said. “Who can guarantee that every single project hits the bullseye on the first shot?”