Bernstein: AI Cost Concerns Overblown, Maintains Outperform on Tencent
N.R. Finch
Bernstein maintains an Outperform rating on Tencent with a HK$780 target — 69% upside — arguing that the market's AI-inference-cost panic is "grossly inaccurate and overly pessimistic" and that the current valuation discount will fade.
Are AI inference costs really that scary?
The core bear case: AI inference costs will crush margins. Bernstein pushes back — cost explosion requires agentic transaction volumes (AI autonomously executing multi-step tasks) to scale in tandem.
A typical chatbot exchange consumes a few hundred tokens. One agentic transaction averages roughly 50,000 tokens — a gap of over 100×. This means → token costs only spike exponentially when agentic volumes actually ramp; the market is over-pricing near-term cost pressure.
On capex, Tencent's trailing-twelve-month spend is 10.9% of revenue, below Alibaba's 12.3%. In plain terms = Tencent is spending more conservatively, leaving it more headroom relative to operating cash flow to invest further.
Where does Tencent's AI product pipeline stand?
The Hy3 model has improved markedly since its April preview. The Xiaowei assistant is now in grey-launch to users, with positive early feedback. Tencent has also signed agent-to-agent agreements with major Android handset makers.
Tencent's Hy team and its WeChat team have operated independently; management says this avoids co-dependency. The WeChat team uses frontier Chinese models such as DeepSeek, and Yuanbao plus Workbuddy have reached 8–9 million MAU.
Bernstein expects — based on Chinese AI labs' pattern of scaling parameters by 2–2.5× per generation — that an Hy4 preview could land around year-end. That release is the next key validation milestone.
How does AI turn into real revenue?
The report highlights WeChat Mini Programs' multi-trillion-RMB GMV (total transaction value on the platform) as Tencent's core AI monetisation opportunity, with the path running through merchant-side services.
This reflects a monetisation logic distinct from Western tech giants — Western investors have largely given up on consumer-facing AI monetisation, while Tencent already holds a ready-made commercial loop in Mini Programs.
In plain terms = Tencent does not need to build a new ecosystem from scratch. Millions of merchants and transactions already run inside Mini Programs; AI just has to lift efficiency and extract value within that existing base.
An 11× P/E — where is the discount?
Tencent trades at 11–12× forward P/E, well below its 2021-to-present historical average of 17–18×. Bernstein argues the discount reflects concerns that are "reasonable for now but will ultimately fade" — slower game growth and AI-related expense pressure.
FCF/EV yields: Tencent ~5%, NetEase ~10%, Boss Zhipin (BOSS) close to 20%. This means → even without a re-rating, some names are already attractive on cash returns alone.
The sector is down ~16% year-to-date (including the recent bounce), trading at 12.1× 2027E P/E — only marginally above the 2022–2023 historical trough.
Who is favoured — and who draws caution?
Outperform maintained: Tencent (target HK$780), Alibaba, NetEase, JD.com (target US$40, ~8.4× 2026E P/E), and Boss Zhipin. JD and BOSS each hold net cash exceeding 40% of market cap, with management actively buying back shares.
More cautious: Meituan has limited upside after its recent rally. PDD Holdings management continues to avoid capital-return dialogue with investors — viewed as the core obstacle for long-term holders.
Bernstein sees three catalysts for a clearer Tencent re-rating: operating-profit forecasts troughing and turning up, a breakout new game, or an Hy4 release that decisively settles the AI-capability debate — these are the key verification windows ahead.
Content is for reference only, not financial advice.