AI Trends

Tencent President Martin Lau Responds to Heavy AI Investment, Expects Clear Improvement in Financial Trends in 2028

EastmoneyUpdated Nov 19 09:47

AI summary

Tencent President Martin Lau likened investment in AI infrastructure to building factories and expects financial trends to improve markedly in 2028 and beyond as facilities are reused, equipment prices fall, and computing efficiency improves. Tencent’s third-quarter capital expenditure reached 81.433 billion yuan, up 527% year on year, while large computing-power purchases pushed free cash flow to negative 69.1 billion yuan; excluding the related prepayments, free cash flow was 16.9 billion yuan. Computing-power supply constraints may delay the release of the next-generation model in the short term, but Lau said inference is unaffected and that Hy4 Flash can run on domestically produced chips.

Why read this

AI score 73

Tencent explained the rationale for its long-term investment in terms of infrastructure reuse, falling equipment prices, and improved computing efficiency, while detailing the impact of computing-power purchases on cash flow.

Topics

Tags

Body · AI translation

Article Body EM_StockImg_Start

EM_StockImg_End

On November 18, Tencent President Martin Lau likened the current round of AI investment to “building a factory” during the post-earnings conference call: “Our infrastructure investment is primarily intended to launch our AI-native businesses. It is a bit like building a factory for model training and Token production. Once the infrastructure is in place, you can continue using it to train new generations of models and continuously generate Tokens.”

Lau believes Tencent’s financial trends will improve significantly in 2028 because it can continue using the infrastructure it has already built. Moreover, in his view, equipment prices for 2026 and 2027 are currently at elevated levels. Looking ahead to 2028, even if the company needs to procure similar equipment, prices are expected to fall substantially.

More importantly, as model training and inference are carried out in practice, computational efficiency is doubling year after year at an astonishing pace. This means that the same facilities will be able to deliver greater training efficiency and produce more Tokens over time.

“Taking all these factors together, we believe the situation will be much better in 2028 and beyond,” Lau said.

The financial report showed that Tencent had net debt of RMB 13.7 billion in the third quarter of this year, compared with net cash of RMB 58.2 billion in the second quarter, mainly due to substantial computing-power purchases, partly offset by RMB 8.4 billion in net cash inflows generated as investment disposals from its portfolio exceeded new investments.

In the third quarter, Tencent recorded negative free cash flow of RMB 69.1 billion, as RMB 2.6 billion in net cash generated from operating activities was more than offset by RMB 65.1 billion in capital expenditure payments, RMB 4.6 billion in media-content payments and RMB 2.0 billion in lease-liability payments. Operating cash flow included substantial prepayments for computing-power purchases to support upgrades to the Hy series of models, WorkBuddy’s inference requirements and the development of AI capabilities across various products and services, while also meeting external customers’ demand for cloud services. Excluding prepayments for computing-power purchases, free cash flow would have turned positive at RMB 16.9 billion. Capital expenditure in the third quarter was RMB 81.433 billion, compared with RMB 12.983 billion in the same period last year, an increase of 527% year on year. Capital expenditure for the first three quarters of this year totaled RMB 166.153 billion, compared with RMB 59.566 billion in the same period last year, an increase of 179% year on year.

Notably, when asked about the impact of computing-power supply, Lau acknowledged that it is an obstacle for all players in China, causing short-term disruption to training and potentially delaying the release of next-generation models. In the long run, however, the domestic supply chain will ultimately mature, while inference remains unaffected. Efficient models such as Hy4 Flash can run entirely on domestically produced chips. “The models we build are themselves extremely efficient, so the domestic supply chain is fully capable of supporting them. There will be no problem.”

Regarding how returns on AI investment should be measured, Lau said the first criterion is whether those investments can create valuable businesses and products, with financial figures coming second. He disclosed that before the Hunyuan Hy large-model series was released, the company was uncertain internally whether it could produce valuable models, but users have now demonstrated explosive acceptance through massive Token consumption.

Lau said Yuanbao had previously fallen short of expectations, so spending has now been substantially reduced and the focus shifted toward capability building. Its AI search capabilities have already been made available externally as a service. “Running a business is not like building a house; you cannot expect to see returns within a few weeks,” he said. “Products will bring user scale, revenue and profit in succession. As for exactly when that will happen, no one knows. We are indeed in the midst of an epoch-making technological revolution.”

In addition to AI, Tencent’s management also cited another reason for pressure on profits in its core businesses. A Tencent executive said the first factors were slower overall growth in fintech services and the growth rate of marketing services. Rising component prices were another key factor. Taking DRAM (dynamic random-access memory) as an example, its year-on-year price increase reached around 400% earlier this year. Memory prices have now stabilized, but Tencent’s core businesses will continue absorbing this round of exceptionally high component costs over the next several quarters.

(Article source: The Paper) Middle of Article Embedded Advertisement

Source

Eastmoney