Tencent Holdings released its third-quarter 2026 results on November 18. Revenue rose 8% year on year to 207.336 billion yuan, operating profit fell 7% to 59.147 billion yuan, and profit attributable to equity holders edged up 0.8% to 63.612 billion yuan.
Two of those three figures point in opposite directions. Operating profit fell 7%, while net profit attributable to shareholders rose 0.8%.
The market consensus had called for revenue of 210.240 billion yuan. The actual figure, 207.336 billion yuan, was about 2.9 billion yuan lower, making this a slightly weaker-than-expected set of results.
The real answer lies not in any of those figures, but in another set of numbers from the same announcement: excluding the impact of investment in new AI products, Tencent’s operating profit was 83.3 billion yuan, up 8% year on year, with its operating margin holding steady at 40%.
In other words, nearly every key number in this earnings report comes in two versions.
Capital Expenditure: 65.1 Billion Yuan or 81.4 Billion Yuan
The difference in the first pair comes down to accounting treatment.
The Shanghai Securities News, Securities Times and Nanfang Plus all reported that Tencent made 65.1 billion yuan in capital expenditure payments during the third quarter, up 226% year on year. China Securities Journal put third-quarter capital expenditure at 81.433 billion yuan, mainly for IT infrastructure, data centers, land-use rights, office campuses and intellectual property.
The 16.3 billion yuan gap between 65.1 billion yuan and 81.4 billion yuan is real. The former is the cash actually paid during the quarter; the latter is the capital expenditure incurred. The difference reflects unpaid liabilities, advance payments and the timing of project acceptance. They are cash-basis and accrual-basis views of the same spending.
In the past, no one paid much attention to this gap. It matters this quarter because the 65.1 billion yuan figure is itself so large: payments jumped 226% year on year, and Tencent spent more in a single quarter than many companies generate in annual revenue. The 81.4 billion yuan incurred figure also means some projects are already underway but have not yet been fully paid for.
Tencent explained where the money went: large advance payments for computing power procurement to support upgrades to the Hy model series, WorkBuddy’s inference demand, and the development of AI capabilities across its products and services.
Research and development spending reached 32.36 billion yuan over the same period, up 42% year on year. Combined with the capital spending, the total approached 100 billion yuan.
Free Cash Flow: Negative 69.1 Billion Yuan or Positive 16.9 Billion Yuan
The second pair diverges even more sharply: one figure is negative and the other positive.
Shanghai Securities News reported that the 65.1 billion yuan in capital expenditure directly pushed Tencent’s free cash flow to negative 69.1 billion yuan, while its balance sheet moved from 58.2 billion yuan in net cash in the previous quarter to 13.7 billion yuan in net debt.
Nanfang Plus reported that free cash flow would have been 16.9 billion yuan excluding advance payments for computing power procurement.
For the same company in the same quarter, free cash flow was either negative 69.1 billion yuan or positive 16.9 billion yuan, depending on whether that advance payment was included. The difference implies that roughly 86 billion yuan was excluded. That is my calculation based on the gap; the announcement did not provide the figure directly.
What does an 86 billion yuan advance payment mean? Tencent effectively pulled several years of future computing power bills forward into this quarter. It is a major hit to cash flow, but it locks in future costs. If computing power prices continue to rise, the advance payment will prove advantageous. If prices fall next year, Tencent will have paid too much too soon.
The shift from net cash to net debt is also worth noting. Tencent swung from positive 58.2 billion yuan to negative 13.7 billion yuan in a single quarter, a movement of 71.9 billion yuan. Tencent said it has ample cash reserves and an investment portfolio worth hundreds of billions of yuan. That is true, but it answers the question of whether the company can absorb the spending, not whether the money was well spent.
Operating Profit: 59.1 Billion Yuan or 83.3 Billion Yuan
The third pair is where the report’s framing is most deft.
The reported figure was 59.147 billion yuan, down 7% year on year. Excluding investment in new AI products, the figure was 83.3 billion yuan, up 8%. The difference suggests that new AI products consumed about 24.2 billion yuan in operating profit during the quarter. That is my arithmetic, not a figure reported by Tencent.
For a company of Tencent’s scale, 24.2 billion yuan is not crippling. But it reveals something important: AI is still purely a cost item, with revenue yet to outpace investment.
There is nothing improper about this disclosure from an accounting standpoint. Tencent clearly states how much it invested and how much it excluded, allowing investors to add the amount back themselves. But the narrative is carefully constructed. “Profit fell 7%” and “profit would have risen 8% without AI investment” describe the same reality, yet sound entirely different.
I do not consider this window dressing. Tencent is more transparent in laying out both measures for the market than it would be in concealing them. But one issue bears watching: whether this exclusion becomes a long-term reporting convention. If Tencent cites growth “excluding AI investment” for four consecutive quarters, it will have shifted from a one-off explanation to a new baseline. Once the baseline moves, comparisons lose their meaning.
What Does 100 Trillion Tokens Mean?
One of the most easily overlooked figures in the report, and one of the most revealing, is WorkBuddy’s consumption.
According to the announcement, WorkBuddy had more than 10 million daily active users as of early November and processed more than 100 trillion tokens per day on average, making it the AI product with the highest daily large language model token consumption in China. Pony Ma used the same description in his remarks.
Divide one figure by the other, and the result is 1 million tokens per user per day. That is my calculation; the announcement did not present it that way.
How much is 1 million tokens? Using an approximate ratio of one Chinese character to 1.5 tokens, that equates to nearly 600,000 to 700,000 Chinese characters. No one could read that much in a day, much less produce it through conversation. There is only one plausible explanation: WorkBuddy is handling tasks, not simply answering questions. A single task might involve reading a batch of files, calling tools several times and producing a draft, consuming tens or even hundreds of thousands of tokens in the process.
That is the dividing line between an agent and a chatbot. A chatbot’s consumption is constrained by how quickly people can type; an agent’s consumption depends on the complexity of the task. Ten million daily active users multiplied by 1 million tokens per user per day suggests WorkBuddy is already being used as a productivity tool, not merely tested as a novelty.
Cloud services revenue, meanwhile, rose 35% year on year. Tencent attributed the growth to two factors: external customer demand for GPU leasing and model as a service, or MaaS, and computing power consumed by its own AI products. The latter is driven by products such as WorkBuddy consuming computing power; the former involves selling that computing power to customers.
Cloud Grew 35%, but Its Broader Segment Accounts for Only 30% of Revenue
Against Tencent’s overall growth rate of 8%, cloud’s 35% increase is more than four times as fast and looks encouraging.
The caveat is the size of the base. Fintech and Business Services revenue rose 9% year on year to 63.655 billion yuan, with cloud services accounting for only a subset of that segment. The entire segment represented about 30% of Tencent’s 207.336 billion yuan in total revenue, and that figure also includes fintech businesses such as payments and wealth management platform Licaitong. Cloud alone accounts for an even smaller share.
In other words, the revenue growth generated by AI is real, but its current scale is not yet large enough to support 65.1 billion yuan in quarterly spending.
Tencent’s other businesses provide useful context. Value-Added Services revenue rose 3% year on year to 98.326 billion yuan. Within that segment, domestic games revenue increased 11% to 47.6 billion yuan, international games revenue fell 9% to 18.9 billion yuan, and social networks revenue declined 1% to 31.8 billion yuan. Marketing Services revenue grew 19% to 43.049 billion yuan. This is where AI is delivering the fastest results: Tencent said the full deployment of its RankUp AI advertising recommendation model across multiple scenarios lifted the corresponding advertising revenue by 2.1% to 4.8%.
Gross margin fell to 56% from 58% in the previous quarter. Gross profit rose just 7% year on year, lagging revenue growth. Depreciation on computing power infrastructure is now flowing into costs, another channel through which AI investment affects the financials. It appears in gross profit rather than operating expenses.
Across the first nine months, cumulative revenue rose 9% year on year to 608.579 billion yuan, while net profit attributable to shareholders increased 7% to 177.727 billion yuan. The quarterly decline in operating profit has not yet become a trend in the year-to-date figures, but it is already moving in that direction.

Comments
00No comments yet. Be the first to weigh in.