After a brief show of restraint in Q1, Tencent went all in on spending in Q2. The company reported after market close that capital expenditure surged to RMB 52.8 billion, about 26 per cent of total revenue and well above the RMB 32.14 billion the market expected. Add RMB 51.4 billion in prepaid compute-procurement deposits in the same period, and Tencent’s effective capex for the quarter reached the hundred-billion level.
The figure set a Tencent record and stands as a statement of intent. For context, Q1 capex was RMB 31.9 billion, up 16 per cent and already a single-quarter high, while last year’s full-year spend was RMB 79.2 billion. Alibaba and ByteDance each spend roughly RMB 120 billion and RMB 160 billion on AI procurement annually.
The spending dragged Tencent’s free cash flow to a negative RMB 13.8 billion, the first negative print since listing, prompting president Martin Lau to explain that AI investment sits mainly in infrastructure and could be leased out through Tencent Cloud even if returns disappoint. The next day Hong Kong shares fell 4.46 per cent, valuing the company at HK$4 trillion, about RMB 3.44 trillion, ceding the top spot to ChangXin Memory.
Tencent’s AI progress has accelerated this year: the Hy3 model, released on 6 July, matched larger models at smaller parameter scale, and WorkBuddy has spawned multiple agent products. WeChat reached 1.439 billion users, up 7 million quarter on quarter, and QQ 520 million, up 4 million. Marketing services revenue was RMB 43.565 billion, up 22 per cent, about 2.7 times the industry average.
The company disclosed that models, Yuanbao, Xiaowei, WorkBuddy and others produced a combined net loss of RMB 10.5 billion. Strip out new AI products and Q2 non-IFRS operating profit would have risen 19 per cent to RMB 86.1 billion. Cash and equivalents stood at RMB 206.93 billion, up 46.7 per cent from end-2025. Chief strategy officer James Mitchell said the top capex priority is training a larger, better Hunyuan model, with the second being inference for Hunyuan and models like DeepSeek behind WorkBuddy.
Per Analysys, Tencent leads China’s AI office-agent market, with WorkBuddy top in traffic ahead of ByteDance’s TRAE IDE and Alibaba’s QoderWork combined. Cloudflare data shows AI-agent traffic on the internet has passed human traffic at 57.4 per cent versus 42.6 per cent, three years earlier than expected.
Editor’s note: This is an adapted translation of the original Sohu report. It has been trimmed and restructured for readability for an international business audience.