← BACK TO FEED
TencentAI infrastructureHunyuanChina techcapital expenditure

Tencent Is Sitting on a $53B AI Money Printer — and Deliberately Not Turning It On

Tencent is forgoing immediate profits from its $53 billion capital expenditure on computing infrastructure, choosing instead to invest in building its own AI models and applications rather than renting capacity to third parties, despite being able to earn over 30% profit margins by doing so. The company believes developing state-of-the-art models like its Hunyuan series and AI products such as WorkBuddy and CodeBuddy will generate greater long-term economic returns than operating as a cloud rental provider. Despite strong Q2 results — including $30.3 billion in revenue and $10.3 billion in net profit — investors reacted cautiously, with Tencent's share price falling around 3% following the earnings announcement.

Tencent spent $53 billion on hardware in a single quarter. It could, apparently, rent that infrastructure out tomorrow and turn an immediate profit. It's choosing not to.

During the company's Q2 earnings call, company president Martin Lau confirmed that demand for compute is so ferocious right now that Tencent has received offers at over 30 percent margin above what it paid for the hardware just months ago. Chief Strategy Officer James Mitchell put it more bluntly: depreciation costs could be recovered 'almost immediately' if they went the neocloud route.

They're not going the neocloud route.

Instead, Tencent is pumping the majority of that compute into building its own frontier models and embedding AI into its own products, with the long-term goal of owning market-leading AI applications in China. Lau described it as 'playing a different game' — betting that superior AI capability converts into superior returns over time, rather than taking the easy rental income now.

The commercial products underpinning that bet include WorkBuddy, an agent platform that Tencent claims can handle complex multi-step tasks end-to-end autonomously, and CodeBuddy, a code generation tool that Mitchell says is actively accelerating cloud migration work — which conveniently creates more revenue for Tencent's cloud business. Vertical integration as a strategy, basically.

On the model side, Tencent released Hunyuan-3 in July — a 295-billion parameter open-weight model. Lau promptly called it 'a very small model,' which is either false modesty or a hint about what's coming. Hunyuan-4 is in the pipeline, and Lau claims it will outperform larger models from competitors. He also mentioned that Tencent's own products are being designed specifically to run on Hunyuan-4, so the model and the applications will be co-optimised. A fifth version is apparently already planned.

The underlying business, meanwhile, is doing fine without any of this paying off yet. Q2 revenue hit $30.3 billion, up 11 percent year-on-year. Net profit came in at $10.3 billion, a nine percent rise. Weixin and WeChat combined added seven million monthly active users to reach 1.349 billion. Advertising revenue jumped 22 percent. Gaming grew 17 percent in China alone.

Investors, however, seem unconvinced. The share price has slid roughly three percent since the earnings announcement. Apparently 'we could make money now but we're choosing not to' is a harder sell than it sounds, even with those underlying numbers.

READ NEXT
US Bans Foreign Robots: Who Gains, Who Gets Hurt, and Why the Drone Precedent Should Worry EveryoneNadMesh Botnet Is Raiding Exposed AI Services for Cloud Keys, and the Numbers Don't Add UpIBM's Mainframe Business Takes a 25% Stock Hit as Customers Blow Budget on AI Hardware