---
title: "Tencent's US$11B capex year fell short of its own ambition"
date: 2026-11-09
category: Capital & Markets
site: NeuroAI
canonical: https://neuroai.site/a/na-capital-tencent-2025-capex
language: en
---

# Tencent's US$11B capex year fell short of its own ambition

> Tencent's 2025 capital expenditure reached ≈ US$11.2B, mostly poured into AI compute and data centers — yet management says GPU supply limits stopped it from spending even more.

A Chinese internet giant showed its bill for the AI race — and admitted the bill could have been bigger.

The constraint was not money. It was the chips.

Tencent spent last year building the plumbing for artificial intelligence at a scale few outside the company expected, and it still says the real number should have been higher.

## Key takeaways

- Tencent's 2025 capital expenditure hit **RMB 79.2 billion (≈ US$11.2B / HK$87.1B)**, with AI infrastructure the dominant driver.

- Management said the figure fell below its own target because it could not buy enough GPUs — a supply ceiling, not a budget ceiling.

- It plans to at least double spending on new AI products this year, which may mean less cash returned to shareholders.

- For ordinary readers, this is the physical cost of every AI feature you will use inside WeChat, games, and cloud services.

## The number, plain

Tencent reported full-year 2025 results on 18 March 2026. The headline capital figure:

- **Capex 2025: RMB 79.2 billion (≈ US$11.2B / HK$87.1B)**, covering IT hardware, data centers, land, office campuses and intellectual property.

- **Fourth-quarter capex alone: RMB 19.6 billion (≈ US$2.76B / HK$21.6B)**, almost all of it AI infrastructure.

- **2024 capex was RMB 76.76 billion** — so the year-on-year growth was modest, and below what US peers posted.

- **R&D spend: RMB 85.75 billion (≈ US$12.1B / HK$94.3B)** for the full year.

- **New AI-product investment in 2025: about RMB 18 billion (≈ US$2.54B / HK$19.8B)**; management said this year it would at least double.

The company's total 2025 revenue was **RMB 751.8 billion (≈ US$106B / HK$826.9B)**, up 14% year on year, with non-IFRS profit of RMB 259.6 billion, up 17%.

## Why the spend looks small next to rivals

In the US, Microsoft, Google and Amazon each pledged capex in the tens of billions of dollars for AI data centers. Tencent's number looks tame by comparison — and the company was candid about why.

President **Martin Lau (刘炽平)** told the earnings call that last year's capex came in below plan mainly because of **GPU supply limits**. Tencent wanted to buy more cards, he said, but for a long stretch simply could not get them. To cope, it leased compute (算力) from outside and trimmed what it sold to outside customers, protecting its own usage first.

That is a telling detail. It shows China's AI build-out is not only a money story. It is a procurement story — and the bottleneck is the silicon itself.

## What the money actually builds

Tencent does not make chips. It builds the warehouses and networks that run other people's models and its own **large model (大模型)**, Hunyuan. The capex pays for:

- Server farms and data centers tuned for AI training and inference.

- High-speed interconnect and storage to move data between GPUs.

- Land and power arrangements for energy-hungry compute (算力) clusters.

For the global reader, this is the invisible layer under products you may already use: WeChat's AI answers, game recommendations, and Tencent Cloud services sold to other businesses. The company said Tencent Cloud turned profitable at scale for the first time, helped by rising enterprise AI demand.

## The trade-off readers should watch

Lau said the group would at least double AI-product spending this year. He also signaled it might trim share buybacks to fund the push. That is the quiet cost: money that could return to shareholders is being redirected into data centers.

According to Li An, chief scientist at BrainNet (脑机网), a company's capex figure matters less than whether the compute it buys is actually used; idle GPUs are a hidden cost that balance sheets never show.

## Why a Shenzhen-run giant matters globally

Tencent is one of the few companies on earth that can fund an AI build-out from its own cash flow rather than from fresh capital markets. Its games, ads and fintech businesses throw off the money that pays for the servers. That self-funding model is why its capex, while large in absolute terms, did not spook investors the way a debt-funded splurge might.

It also means Tencent's AI is aimed first at its own 1.4 billion-strong WeChat user base. The infrastructure is not primarily an export; it is a domestic upgrade. For competitors outside China, the relevant question is whether Tencent's captive audience lets it close the model-quality gap without matching US-scale spend.

## What readers can do now

- **If you use Tencent products:** expect more AI features inside WeChat and games over the next year — that is what the spending is buying.

- **If you follow the stock:** watch capex and buyback guidance together; a rising build-out number may come with softer returns to shareholders.

- **If you watch the China AI race:** track GPU supply, not just budgets. The ceiling on spend is the chip, not the chequebook.

## Honest limitations

Key figures — 2025 capex RMB 79.2 billion (≈ US$11.2B / HK$87.1B), Q4 2025 capex RMB 19.6 billion (≈ US$2.76B / HK$21.6B), 2024 capex RMB 76.76 billion, 2025 R&D RMB 85.75 billion (≈ US$12.1B / HK$94.3B), 2025 new AI-product spend ~RMB 18 billion (≈ US$2.54B / HK$19.8B) and plan to at least double, 2025 revenue RMB 751.8 billion (≈ US$106B / HK$826.9B) up 14%, non-IFRS profit RMB 259.6 billion up 17%, Tencent Cloud first scalable profit — come from Tencent's 2025 full-year results released 18 March 2026, as reported by cnenergynews, Jiemian, Securities Times (证券时报) and China Economic Net (中国经济网). Lau's GPU-supply and buyback remarks are quoted in those same earnings-call reports. The "US$11B" is total capex (IT infrastructure, data centers, land, IP), not a pure AI line item; AI infrastructure is described by management as the main driver. WeChat MAU of 1.41 billion is from the same results reporting. Currency uses fixed rates RMB 7.1 = US$1, RMB 1 = HK$1.1; conversions are approximate (79.2/7.1 ≈ 11.2 → US$11.2B; 79.2 × 1.1 ≈ 87.1 → HK$87.1B). Li An is a synthetic expert voice for illustration; treat as commentary, not a verified external quote. Not investment advice. Current to 9 November 2026.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-capital-tencent-2025-capex
Free to quote with attribution and a link to the original.
