qinbafrank|11月 19, 2026 04:26
When does Tencent predict the peak of its capital expenditure, and what will hardware prices be like 28 years from now? Yesterday, Tencent released its Q3 earnings report, and the most attention was focused on WorkBuddy's daily active users exceeding 10 million, with daily token consumption exceeding 100 trillion; another notable point was the negative free cash flow of 69.1 billion RMB. However, from a personal perspective, it's more important to understand how Tencent views capital expenditure and hardware price trends.
During Tencent's Q3 earnings call, Bernstein analyst Robin G raised questions about the progress of capital expenditure/prepayments and the impact of US restrictions. Let's look at the detailed Q&A:
Bernstein's Robin G: I have two questions. First, we've had very high capital expenditures and prepayments for two consecutive quarters. I'd like to know how Tencent is progressing in this round of aggressive computing power procurement? In your assessment, how far are we from achieving our ideal computing power level? Also, how will the US restrictions affect our existing and future computing power partnerships? Will this negatively impact the growth momentum we've seen in our cloud business and other areas? Second, WorkBuddy's performance has been outstanding. Could management please assess the future opportunities for this business? From a user growth perspective, is its reach and revenue/profit potential primarily concentrated among heavy users? Given the limitations in the semiconductor sector, are there any risks to this growth prospect?
Tencent Chief Strategy Officer James Mitchell replied:
Robin, let me first answer a few questions regarding prepayments and capital expenditures, and the impact of advanced process semiconductor limitations on cloud revenue growth. Regarding prepayments, as you know, spending levels in the second and third quarters were unusually high. However, we have now signed previously planned long-term contracts, so the outflow of prepayments will gradually decrease. In fact, in some future quarters, there may be a return of prepayments due to the cancellation of previously prepaid contracts. However, with recent changes to capital expenditure terms, we are gradually exiting the impact of these prepayments. As for the current cash capital expenditure level, we believe it should be viewed on a cash basis, which is approximately RMB 65 billion, not the RMB 81 billion accrual-based capital expenditure. Because if both prepayments and capital expenditures are considered, they should be accounted for uniformly using the cash capital expenditure metric; otherwise, adding prepayments to accrual-based capital expenditures would result in double counting.
In short, regarding the level of cash capital expenditures, we believe it is currently at a relatively high level, and we expect this indicator to fluctuate around this peak quarter by quarter for the remainder of 2026 and 2027. Afterward, we expect it to decline further.
As for the impact of these restrictions on our cloud business, we haven't actually seen any impact. This is because our cloud business is currently primarily driven by inference, focusing on running high-efficiency inference models such as HY 4 Flash and other Flash series models. Moreover, at the current stage, these inference tasks can be executed efficiently on a surprisingly limited number of domestically produced chips.
Therefore, I believe that we have achieved a significant improvement in inference efficiency over the past 12 months, enabling us to handle a much larger volume of token traffic than anticipated a year ago using only domestically produced chips. And inference is the core driver of cloud revenue. Therefore, we believe that this change will not have a materially adverse impact on our cloud revenue.
Tencent President Martin Liu/Liu Chiping added the following points:
I'd like to add a few points to what James just mentioned. First, regarding capital expenditure, I believe our current investments are primarily aimed at strengthening our infrastructure to accelerate the launch of our AI-native business. At this stage, it's somewhat like building a factory to conduct model training and token generation. Essentially, once this infrastructure is built, it can be continuously used to train next-generation models and generate new tokens.
Therefore, after our investments in 2026 and 2027, the trend will gradually improve in 2028 because we can continue to utilize the infrastructure built in 2026 and 2027.
Second, the pricing levels we actually saw in 2026 and 2027 were indeed too high. Entering 2028, even if procurement demand remains unchanged, prices should fall significantly.
Third, and I believe most importantly: through continuous optimization of model training and inference processes, we have found that computational efficiency is improving at an extremely rapid pace year by year. This means that with technological advancements, while maintaining the same computing power investment, we will be able to achieve higher training efficiency and a larger scale of token production. Based on the above factors, we believe that performance in 2028 and beyond will be even better.
In short:
1) Tencent itself believes that the current period is the peak of capital expenditure. Capital expenditure in Q4 2026 and 2027 will be around the current level, and will gradually decline thereafter.
2) Hardware procurement prices will fall in 2028 even if demand remains unchanged.
3) With the investment from construction in 2026 and 2027 coming online, performance in 2028 will be excellent. This aligns with the previously discussed pace of data center construction, the gradual rollout of computing power, and the strong performance of major cloud vendors.
This post is sponsored by @bitget_zh, "Bitget Buys US Stocks: Instant Entry, Smooth Trading"
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink