Tencent’s Capex to Remain at Peak Levels Through 2027 on AI Investment(Yicai) Nov. 19 -- Tencent Holdings’ capital expenditure will remain at peak levels through next year as a result of heavy investment in artificial intelligence infrastructure, but the Chinese internet giant anticipates an upswing in financial performance from 2028 onward, executives said on its latest earnings conference call.
Capex prepayments were unusually high in the second and third quarters of this year, Chief Strategy Officer James Mitchell said on the call late yesterday, adding that as Tencent has already secured long-term contracts, cash outflows for prepayments will decline and some quarters may even see reversals.
Cash-based capex, these prepayments, has peaked and is likely to fluctuate around peak levels for the rest of this year and through 2027, before starting to decline, he noted.
Capex surged more than six times to CNY81.4 billion (USD12.2 billion) in the three months ended Sept. 30 from a year earlier, with cash payments of CNY65.1 billion, the Shenzhen-based company’s third-quarter earnings report showed.
President Martin Lau said current infrastructure investments are mainly aimed at launching AI-native businesses, and once this infrastructure in in place, it will enable the continuous training of new-generation models and generate tokens. So when capex commitments for this year and the next are completed, Tencent will be able to use the infrastructure in 2028, leading to a marked improvement in its financial performance, he pointed out.
During Tencent’s recent annual management meeting, its leadership emphasised that if the company fails to get on board fully with the AI revolution, then it risks being disrupted and losing ground to competitors, according to the minutes posted on its social media accounts on Nov. 16. Tencent wants half of its 100,000 employees to act as an “army,” defending existing businesses, while the other half acts as a “navy,” exploring new opportunities in AI.
Equipment prices have been elevated this year and will remain so in 2027, but are expected to fall substantially in 2028, Lau said on the call, adding that in the course of AI model training and inference, computing efficiency is doubling year by year, meaning that the same facilities will deliver higher training efficiency and greater token output in the future.
Based on non-international financial reporting standards, third-quarter operating profit fell 6 percent to CNY68.1 billion from a year ago, while revenue rose 8 percent to CNY207.3 billion (USD31.1 billion), according to the earnings report.
Tencent’s shares [HKG: 0700] finished 6.6 percent lower at HKD348.40 (USD44.41) each today, bringing the stock’s decline to 42 percent since the end of last year. The wider Hong Kong market lost almost 2 percent today.
Chip-Buying Curbs
Investment in AI infrastructure involves buying advanced-type chips, so Tencent’s management also addressed the impact of supply restrictions on the call.
The restrictions are a hurdle for all domestic players, and whether they rent or build computing power overseas, the advanced compute available has been reduced, Lau noted. Training will be affected, so Tencent’s next-generation model may take longer to release, he added.
But in the long run China’s supply chain is expected to mature and catch up, and on the AI inference side, since the models are highly efficient, the domestic supply chain is fully capable of supporting them, he pointed out. All Chinese players will likely eventually rely on domestically produced supply chains and chips, he stressed.
According to Mitchell, domestic chips can already run inference workloads efficiently on highly efficient models such as Tencent’s Hy4 flash. Over the past 12 months, inference efficiency has undergone a major shift, with token throughput on domestic chips far higher than anticipated a year ago, he said, noting that Tencent continues to add computing capacity through the Chinese supply chain.
Editor: Martin Kadiev
