Tencent president expects AI spending pressure to ease by 2028

In a November 18, 2026 earnings call, Tencent president Martin Lau Chi-ping compared the company’s AI investment to building a factory: infrastructure built for model training and token production can continue to serve later generations of models. He said Tencent expects its financial position to improve markedly in 2028 as it reuses that infrastructure. That is management’s forecast, not a reported result.

Lau said equipment prices were high in 2026 and 2027 but expected comparable equipment to cost substantially less in 2028. He also argued that improving computing efficiency would allow the same infrastructure to train models more efficiently and produce more tokens over time.

The spending was already visible in Tencent’s third-quarter figures, as reported on November 19, 2026. The company recorded RMB 13.7 billion in net debt, compared with RMB 58.2 billion in net cash in the second quarter. The report attributed the shift mainly to large computing-capacity purchases, partly offset by RMB 8.4 billion in net cash inflows as disposals from its investment portfolio exceeded new investments.

Third-quarter free cash flow was negative RMB 69.1 billion. Operating cash inflow of RMB 2.6 billion was more than offset by RMB 65.1 billion in capital-expenditure payments, RMB 4.6 billion in media-content payments and RMB 2 billion in lease-liability payments. Operating cash flow included substantial advance payments for computing capacity to support upgrades to the Hy model series, WorkBuddy inference demand, AI features across Tencent products and services, and external cloud customers. Excluding those advance payments, free cash flow would have been positive RMB 16.9 billion.

Capital expenditure reached RMB 81.433 billion in the quarter, up 527% from RMB 12.983 billion a year earlier. For the first nine months of 2026, it totaled RMB 166.153 billion, up 179% from RMB 59.566 billion in the comparable period.

Asked about computing supply, Lau said constraints affected AI developers across China and could delay Tencent’s next model in the short term. He said he expected the domestic supply chain to mature and did not see the same constraint for inference, stating that efficient models such as Hy4 Flash could run entirely on domestically produced chips.

Lau said the first test of AI investment was whether it produced valuable businesses and products, with financial returns coming later. He said Tencent had sharply reduced spending on its Yuanbao app after it fell short of expectations and was focusing instead on building capabilities, including an AI search service made available externally. He did not give a timetable for when AI products would deliver users, revenue or profit.

Management also pointed to pressure on core-business profit beyond AI spending. A Tencent executive cited slower growth in fintech services and the pace of marketing-services growth, as well as higher component costs. The executive said DRAM prices had risen by about 400% year on year earlier in 2026; although memory prices had since stabilized, the company expected to absorb those unusually high costs over the following quarters.