The Market Changed the Yardstick on Tencent

At the Hong Kong close on Oct. 21, Tencent closed at HK$393.8, down 4.04%, with a total market capitalization of HK$3.58 trillion. At the start of the year, the stock traded around HK$633.7, a market value of HK$5.76 trillion. The year-to-date decline is nearly 40%.

Revenue in the first half of 2026 was RMB401.2 billion, up 10% from RMB364.5 billion a year earlier. Profit was RMB117.369 billion, up 11% from RMB105.769 billion. Domestic game revenue grew 6% in Q1. International games grew 13%. Gross billings for Honor of Kings and Game for Peace were still hitting new highs. No substitute has emerged for WeChat’s social graph, and its communications role remains intact.

Operating performance has not broken. The stock has fallen to a near one-year low. The market has changed the measure it uses to value Tencent.

A cash machine starts writing large checks

For years, Tencent looked stable in valuation models: ample net cash, real-cash buybacks every quarter, free cash flow positive year after year. Buybacks reached about HK$112 billion in 2024 and supported the share price. The valuation carried an implicit promise: cash earned would return to shareholders through buybacks and dividends.

In Q2 2026, that promise was revised.

Quarterly capital expenditure was RMB52.78 billion, up 176% year on year and 65% quarter on quarter. First-half capex totaled RMB84.72 billion, nearly double the prior-year period. Free cash flow was negative RMB13.8 billion, a rare quarterly negative for Tencent. Net cash from operations was RMB52.7 billion. Capex payments were RMB59.3 billion, media content payments RMB5.0 billion, and lease liability payments RMB2.2 billion. Excluding prepayments for computing power procurement, free cash flow was RMB37.6 billion. The negative line on the statements was still glaring.

Profit felt it immediately. New AI products (Hunyuan, Yuanbao, CodeBuddy, WorkBuddy and Xiaowei) dragged Non-IFRS operating profit down by about RMB10.5 billion in Q2. UBS estimated AI-related operating expenses at RMB16 billion for the quarter. Adjusted operating profit fell 2% year on year to RMB71 billion. Goldman Sachs raised its 2026 full-year capex forecast from RMB151 billion to RMB210 billion. UBS raised its from RMB170 billion to RMB250 billion. In October, Tencent issued US$1.8 billion and RMB26 billion in notes. The market read that as stockpiling ammunition for more AI investment.

Supporting forces retreated at the same time. Buybacks fell from HK$58.763 billion in the first three quarters of 2025 to HK$31.441 billion, a 46.5% drop. Daily buybacks once bottomed at about HK$100 million, against a 2025 daily average of HK$635 million. South African major shareholder Prosus sold about 13 million shares in Q1 2026, raising nearly HK$7.3 billion. Hong Kong liquidity also tightened. Southbound net inflows were HK$440 billion in the first nine months, far below HK$1.4 trillion for all of 2025.

The chain is easy to trace. AI capex expands. Free cash flow turns negative. Buyback capacity is diluted. Supporting funds exit. The share price loses support. More allocation-driven funds leave. The chain starts with a reversal in expectations about where Tencent’s money is going.

The valuation anchor slides from growth stock to hybrid pricing

Blame “AI spending dragging down profit” and you only reach the financial surface. The pricing framework has changed.

Institutional consensus puts Tencent’s 2026 forward PE at about 15 to 16 times. Strip out the value of external investments from market capitalization, and the pure core business trades at about 13 times. Morgan Stanley was direct after Q2. It expects earnings to be roughly flat from the second half of 2026 through 2027 because of heavier AI investment. It cut its target price from HK$650 to HK$550 and kept “Overweight.” Goldman Sachs cut its target from HK$700 to HK$670 and kept “Buy,” citing pressure on near-term earnings per share from AI investment. It later adjusted the target to HK$650 on “high AI investment and recent equity financing.” Repeated target cuts show the market disagrees on the return on AI investment.

A company whose profit will be roughly flat for two to three years will not command a growth-stock premium. Foreign investors have repriced Tencent from a cash cow with high cash flow and high buybacks into an asset burning cash on AI bets. The PE center fell from 20 times to 15 to 17 times. The contradiction is obvious. A 15 to 17 times PE is a value-stock reading. Utility-stock pricing assumes cash can be steadily distributed. Tencent is reinvesting cash at scale. The result is hybrid pricing. Core social and gaming provide stable cash flow and a 4% to 5% dividend yield. The external investment portfolio provides asset value. Together they support a value-stock floor. The free-cash-flow hit and return uncertainty from AI investment deduct another discount from that floor. The share price roughly equals the value-stock floor minus the AI option discount.

The discount is hard to close because of how revenue is disclosed. AI is pulling revenue. Q1 marketing services revenue grew 20% year on year. Q2 advertising revenue grew 22%, driven by the AI delivery system AIM+ and recommendation model iteration. Cloud revenue grew 40% year on year. AI-related services are already 25% to 30% of cloud revenue. Tencent does not separately disclose direct AI revenue. AI contributions sit inside advertising and cloud line items. Cost-side depreciation, R&D and operating expenses are quantified and magnified. Excluding the drag from new AI product investment, Non-IFRS operating profit grew 19% year on year. Reported growth was 9%. The market sees a clear AI bill on the cost side and a vague AI enablement on the revenue side. Costs are visible. Revenue is not. Pricing biases toward suppression.

The slow shift in attention share

AI investment is the acute pressure on valuation. Attention structure is the quieter erosion.

In July 2026, Douyin’s main app reached a 19.4% share of single-app time, passing WeChat’s 18.8% for the first time. ByteDance’s overall time share was 40.9%. Tencent’s was 29.1%. The gap was widening. Hongguo’s standalone short-drama app reached 168 million daily active users, more than the combined DAU of iQIYI, Youku, Tencent Video and Mango TV. Average daily use was about 125 minutes. WeChat was at about 82 to 85 minutes over the same period. Channels was about 50 minutes.

WeChat’s moat is its social graph and the necessity of communications. That part is not under substantive threat. Users still open WeChat every day. The incremental attention they spend killing time after opening it flows in directions WeChat has not focused on defending. Tencent confines short dramas to a secondary mini-program container, “Green Bubble.” It deliberately keeps them out of WeChat’s home feed to protect WeChat as an acquaintance-social tool. The defensive move is rational. It also shows that WeChat’s DNA is not advantaged in competing for addictive time. The capital market is pricing this structural unease. When the incremental attention pool flows elsewhere, can Tencent capture it with equal efficiency?

Geopolitical and liquidity discounts remain

Beyond those two threads, another factor keeps working. Hong Kong shares have long traded at a 10% to 20% discount to U.S. ADRs. The discount widens in weak markets. Some analyses break Tencent’s discount into four layers: geopolitics, regulatory policy expectations, exchange-rate expectations and Hong Kong liquidity. The combined discount is estimated at 40% to 50%. Meta trades at about 25 times forward PE. Google trades at about 22 times. Tencent’s core business trades at about 13 to 14 times. Company operations cannot close that gap alone. It depends on when pricing power migrates from foreign investors to southbound capital. Southbound net inflows have slowed since 2026. The migration pace is more uncertain.

The historical pattern cannot be extrapolated directly

Tencent has been through six valuation troughs: the 2008 financial crisis, the 2010 3Q War, payment regulation in 2014, rate-hike expectations in 2015, the game license freeze in 2018, and antitrust in 2022. Each time, what the market feared did not ultimately destroy Tencent. That history offers a hedging perspective. It cannot be directly extrapolated to the present. The past six shocks came mainly from external policy or competitive threats. This time, Tencent has chosen to put a large amount of cash flow into an area whose return path is not yet clear.

Morgan Stanley and Goldman Sachs cut target prices but kept “Overweight” and “Buy.” That reflects the split. They recognize the quality of the core assets. They refuse to pay for the growth option before AI returns are visible. Yuanbao’s MAU was 57.35 million. Doubao’s was 345 million. Qwen’s was 166 million over the same period. Pony Ma admitted at the annual meeting that Tencent had been slow on AI, nine months to a year behind. Compute procurement has physical constraints. Because of sanctions, whoever locks up GPUs first has the training and inference capacity. The waiting period missed an irreversible compute window. AI infrastructure investment is continuous. Model iteration, compute expansion and inference costs rise with user growth. Goldman Sachs forecasts capex of RMB210 billion, RMB245 billion and RMB247 billion for 2026 through 2028.

Watch three things in the next earnings report. Can Tencent disclose AI revenue in a separately attributable way? Can free cash flow turn positive? Can buybacks return to the 2025 scale? If those three do not change, the discount stays in the share price.