On Oct. 13, Tencent Holdings began marketing the issuance of U.S. dollar and offshore renminbi bonds, with the offering expected to raise about $5 billion to support its growing investments in artificial intelligence. According to media citing a person familiar with the matter, the Chinese technology and gaming giant has set initial price guidance for its proposed 10-year and 20-year dollar bonds at roughly 115 and 140 basis points over comparable U.S. Treasuries, respectively. The person said the company is also marketing 10-year and 30-year offshore renminbi bonds, with price guidance of around 2.9% and 3.5%, respectively.
The person said the bonds could price as early as Tuesday.
Sources citing Tencent management said the proceeds will be used mainly for general corporate purposes, including AI-related capital expenditure, with the bulk of AI investment going toward chip procurement. Tencent confirmed this allocation on a call with investors. Sources further noted that Tencent's capital expenditure may remain elevated for the rest of this year and into 2027, implying annual spending of slightly above 200 billion yuan based on second-quarter levels.
Tencent to use new bond proceeds to ramp up AI investment
The backdrop to Tencent's intensive financing is a sharp climb in its AI capital expenditure. In the second quarter of 2026, Tencent's capital spending surged to 52.8 billion yuan, far exceeding the expected 32.1 billion yuan and nearly tripling the 19.1 billion yuan recorded in the same period a year earlier. Tencent management said this reflects strategic upfront investment in AI infrastructure to support upgrades to the Hunyuan model, inference demand for WorkBuddy and CodeBuddy, WeChat AI, and growing external customer demand for cloud services.
Tencent President Martin Lau addressed the company's rising capital expenditure on the second-quarter earnings call. He said Tencent is indeed making massive investments in computing power and has already seen clear upside in returns, with multiple new applications performing well. For some of the compute orders placed earlier, if sold they could yield profits more than 30% above the procurement prices from a few months ago. Lau emphasized, "If we advance in the sequence of first using compute for proprietary model training, then for application development, and finally renting out compute capacity externally, we can build a native AI business of tremendous scale, which would bring very substantial profits."