Tencent Holdings Limited (0700.HK) Report Interpretation

Overview

This report evaluates the release timing, technical specifications, model performance, application productivity, and commercial pricing of Tencent’s Hy4 preview and places these developments within Tencent’s existing rating and valuation framework. Morgan Stanley’s core judgment is that Hy4 was released earlier and is more capable, Tencent’s large-model iteration is accelerating, and model performance is now approaching the top tier.

Core views

First, the launch of the Hy4 preview was significantly earlier than Morgan Stanley had previously expected. During its second-quarter earnings release, Tencent stated only that Hy4 would be launched within the year. Morgan Stanley had originally expected the full version to be released in the fourth quarter of 2026 and a preview to appear in September, but Tencent released the preview on August 28, 2026. The report also observes an acceleration in the model iteration cycle: after Yao Shunyu joined in the fourth quarter of 2025 as chief AI scientist, the Hy3 preview was launched in April, the full Hy3 version in July, and the Hy4 preview subsequently debuted at the end of August. Second, the Hy4 preview represents a substantial expansion over Hy3 in model size and long-context capability. It uses a mixture-of-experts architecture, with 770 billion total parameters, 49 billion parameters activated per inference, and a context window of 1 million tokens. Hy3 had 295 billion total parameters, 21 billion activated parameters, and a 256,000-token context window. Morgan Stanley believes that high-quality training data and co-design supported the improvement in model capabilities. Third, model test results indicate that the Hy4 preview is already competitive with the top tier. In Tencent’s internal blind test, the Hy4 preview achieved an average score of 2.99/4, slightly above GLM-5.3’s 2.92/4 and Kimi K3’s 2.94/4. On Arena.ai’s Code Arena: WebDev leaderboard, the Hy4 preview ranked fifth with a score of 1,633. Higher-scoring models listed in the report included Kimi K3 at above 1,674, Qwen-3.8 Max at 1,669, and GLM-5.3 Flash at 1,634. Beyond benchmark testing, the report states that the model achieved notable improvements in practical tasks such as software development, office productivity, game development, and scientific research. Fourth, Tencent simultaneously raised its cloud API fees. Hy4’s prices for cache-hit input, cache-miss input, and output are RMB0.3, RMB6.0, and RMB18.0 per million tokens, respectively, compared with RMB0.25, RMB1.0, and RMB4.0 for Hy3. The report therefore presents not only a technical upgrade but also a higher level of commercial pricing, although it does not separately quantify the incremental contribution to revenue or profit. Regarding the stock view and valuation, Morgan Stanley assigns Tencent an “Overweight” rating and an “Attractive” industry view. The target price is HK$550, implying 21% upside from the closing price of HK$455.20 on August 28, 2026. The target price uses a sum-of-the-parts valuation: the core business is valued at HK$475 using a DCF calculation with a 10% discount rate and 3% perpetual growth rate; associate investments are valued at HK$75 per share based on book value with a 30% discount applied to investment value. The target-price history shows HK$650 on March 19, 2026, adjusted to HK$550 on August 13, 2026. This report does not explicitly announce a new rating or target-price adjustment. The cover financial table presents figures for 2025, 2026 estimates, 2027 estimates, and 2028 estimates: net revenue of RMB751.8 billion, RMB829.4 billion, RMB908.7 billion, and RMB989.6 billion, respectively; EBITDA of RMB307.6 billion, RMB345.6 billion, RMB388.2 billion, and RMB440.9 billion; and ModelWare net income of RMB224.8 billion, RMB229.8 billion, RMB231.3 billion, and RMB233.9 billion. The table shows P/E ratios of 22.1x, 15.8x, 15.7x, and 15.5x; EV/EBITDA of 14.2x, 8.8x, 7.7x, and 6.5x; EV/revenue of 6.8x, 4.5x, 4.1x, and 3.6x; P/BV of 4.3x, 2.7x, 2.3x, and 2.0x; and ROE of 21.2%, 19.2%, 17.6%, and 16.8% over the same periods. Potential upside drivers listed in the report include solid execution of new domestic and overseas game launches, gains in social and short-video advertising share, resilience in social networking and online entertainment amid competition, and acceleration in consumer-facing AI applications. Downside risks include regulatory uncertainty in the gaming industry, intensifying competition for social-network users and advertising budgets from emerging entertainment formats, tighter antitrust regulation, and tensions in China-U.S. relations.

Analysis framework

The report first compares Hy4’s actual release date with Tencent’s previous guidance and Morgan Stanley’s prior expectations, then evaluates the technical upgrade based on parameter scale, activated parameters, and context window. It subsequently conducts cross-validation using Tencent’s internal blind test, Arena.ai rankings, and practical task performance, while also examining changes in Tencent Cloud API pricing. Finally, the report connects product developments with equity valuation through ModelWare financial metrics, DCF, and sum-of-the-parts valuation, while listing operating and regulatory factors that could support or constrain realization of its view.

Methodology notes

• Valuation MethodSOTP Valuation

  Sum-of-the-Parts Valuation

  The report divides Tencent into its core business and associate investments, values each separately, and adds them together to derive a base-case target price of HK$550.

• Valuation MethodDCF

  Discounted Cash Flow Valuation of Core Business

  The HK$475 core-business value is calculated using DCF, applying a 10% discount rate and 3% perpetual growth rate to discount future cash flows to present value.

• (Out-of-Vocabulary Method)

  Benchmark Testing and Internal Blind Testing

  The report uses Tencent’s internal anonymized comparison scores and Arena.ai’s public leaderboard to compare the Hy4 preview horizontally with models such as GLM, Kimi, and Qwen and determine whether its capabilities approach the top tier.

• (Out-of-Vocabulary Method)

  Morgan Stanley ModelWare

  The report states that its cover financial metrics are primarily based on Morgan Stanley’s ModelWare framework and uses it to present revenue, EBITDA, net income, and valuation multiples.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Tencent Holdings Limited (0700.HK)

  The early release of the Hy4 preview, improved model capabilities, and higher cloud API pricing reinforce the report’s positive assessment of Tencent’s progress in AI iteration.

  Strengths

  Accelerating model iteration; Hy4’s internal blind-test score was slightly higher than GLM-5.3 and Kimi K3; notable improvements in software, office productivity, game development, and scientific research tasks; positive operating drivers in gaming, advertising, and social businesses.

  Weaknesses

  The report does not separately quantify Hy4’s contribution to revenue, profit, or cash flow.

  Comparison

  Hy4 has 770 billion total parameters, 49 billion activated parameters, and a 1 million-token context window, all higher than Hy3. Its internal blind-test score was slightly above GLM-5.3 and Kimi K3, but its Arena.ai web development coding score was lower than the Kimi K3, Qwen-3.8 Max, and GLM-5.3 Flash scores listed in the report.

  Risks

  Regulatory uncertainty in the gaming industry, intensifying competition for social-network users and advertising budgets, tighter antitrust regulation, and tensions in China-U.S. relations.

Key data

• Hy4 Preview Release DateAugust 28, 2026Earlier than Morgan Stanley’s original expectation of a September preview
• Model Parameters770 billion total parameters, 49 billion activated parametersHy3 had 295 billion total parameters and 21 billion activated parameters
• Context Window1 million tokensHy3 had 256,000 tokens
• Tencent Internal Blind Test2.99/4GLM-5.3 scored 2.92/4, while Kimi K3 scored 2.94/4
• Arena.ai Web Development Coding LeaderboardFifth place, 1,633 pointsThe report lists Kimi K3 at above 1,674 points, Qwen-3.8 Max at 1,669 points, and GLM-5.3 Flash at 1,634 points
• Tencent Cloud Hy4 API PricingRMB0.3/6.0/18.0 per million tokensRespectively corresponding to cache-hit input, cache-miss input, and output; Hy3 was priced at RMB0.25/1.0/4.0
• Rating and Industry ViewOverweight; AttractiveThe stock rating is Overweight, and the industry view is Attractive
• Target Price and UpsideHK$550.00; 21%Based on the HK$455.20 closing price on August 28, 2026
• Target Price CompositionCore business HK$475 + associate investments HK$75/shareThe core business uses a DCF with a 10% discount rate and 3% perpetual growth rate; a 30% discount is applied to investment value
• Recent Target Price HistoryReduced from HK$650 to HK$550The historical records correspond to March 19, 2026 and August 13, 2026, respectively
• Net RevenueRMB751.8/829.4/908.7/989.6 billionFor 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively
• EBITDARMB307.6/345.6/388.2/440.9 billionFor 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively
• ModelWare Net IncomeRMB224.8/229.8/231.3/233.9 billionFor 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively
• EPS (RMB, ** basis)24.32/24.76/24.92/25.21For 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively; the report notes that ** is based on the consensus methodology
• EPS (RMB, $ basis)27.99/29.64/31.70/34.27For 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively; the report notes that $ denotes Refinitiv Estimates consensus data
• Valuation MultiplesP/E 22.1/15.8/15.7/15.5x; EV/EBITDA 14.2/8.8/7.7/6.5xFor 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively
• Other Valuation and Return MetricsEV/revenue 6.8/4.5/4.1/3.6x; P/BV 4.3/2.7/2.3/2.0x; ROE 21.2%/19.2%/17.6%/16.8%For 2025, 2026 estimates, 2027 estimates, and 2028 estimates, respectively
• Market Data52-week range HK$683.00–411.00; diluted shares 9,279 million; market capitalization US$538,771 million; enterprise value US$458,191 millionCurrent market metrics shown on the cover; average daily trading value was US$1,723 million

Impact & implications

Morgan Stanley views the earlier release, expanded model scale, near-top-tier test performance, and improved productivity in practical tasks as evidence that Tencent’s AI R&D iteration is accelerating. Tencent Cloud’s API price increase indicates that the new model carries higher pricing, but the report does not quantify its revenue contribution. For the stock view, AI progress and potential positive factors in gaming, advertising, and social businesses jointly support the “Overweight” stance, while the target price remains dependent on the DCF value of the core business, the discount applied to investment assets, and related operating and regulatory risks.

Risks

• The gaming industry continues to face regulatory uncertainty.
• Emerging entertainment formats may intensify competition in social networking and for advertising budgets.
• Tighter antitrust regulation and tensions in China-U.S. relations may pose downside risks.