Turning Point in the AI War? Wall Street's Takeaway After Meeting Management: TENCENT Is Fighting Its 'Battle of Midway'

TENCENT, long viewed as a chaser in the AI arms race, is approaching a potential strategic turning point.

According to Zhuifeng Trading Desk, Bank of America Merrill Lynch noted in its latest report that it organized an investor roadshow in Hong Kong following the release of TENCENT's second-quarter 2026 results. The event featured face-to-face exchanges between institutional investors and TENCENT's Chief Strategy Officer Jame Mitchell, Chief Financial Officer John Lo, and the investor relations team.

The report stated that investor questions were almost exclusively focused on AI strategy—capital expenditure outlook, AI monetization paths, dynamics in the computing power market, and model progress. Following the meeting, the firm maintained its "Buy" rating on TENCENT and its HK$780 target price, characterizing this moment as the "Battle of Midway" in TENCENT's AI competitive landscape—the war is not yet over, but the balance of power has begun to shift quietly.

Bank of America Merrill Lynch analysts Alex Liu and Joanna Du pointed out in the report that TENCENT's rapid catch-up in model capabilities and computing power accumulation within the year marks a potential inflection point in the long-term competitive landscape. Although the market currently remains focused on leaders in the "cloud computing/chip" sector, TENCENT's AI option value (including upside potential in token and cloud revenue) and the resilience of its core business are not fully reflected in its current valuation, which stands at less than 13 times forward P/E for 2026/2027. TENCENT's stock is currently trading at HK$440, leaving more than 77% upside to the target price of HK$780.

Computing Power Layout: Strategic Asset, Not Just Cost

Computing power construction was one of the core topics of this investor meeting. Management stated that TENCENT significantly increased capital expenditure on AI infrastructure in the second quarter of 2026 and made substantial advance payments to lock in supplies of computing power and memory before industry prices rose. Management noted that the advance payments were primarily used to secure memory and other computing resources, with most capital expenditure currently deployed domestically.

Based on this, the report raised its capital expenditure forecasts for TENCENT in 2026 and 2027 to RMB 210 billion and RMB 260 billion, respectively (previously forecast at RMB 185 billion and RMB 225 billion). However, analysts believe that quarterly cash pressure from computing power investments will peak in the second and third quarters of 2026. Thereafter, as non-recurring expenses related to advance payments gradually normalize, operating cash flow is expected to stabilize in 2027.

Management emphasized that TENCENT views ownership of computing power as a strategic asset, whose value is reflected not only in cloud monetization but also in supporting the competitiveness of foundation models, AI applications, and the broader ecosystem. TENCENT also possesses the flexibility to dynamically allocate computing resources across cloud customers, model training, and inference workloads, scheduling them based on demand and expected returns.

Hunyuan Model: Trading Inference Efficiency for Ecosystem Scale

In terms of model strategy, TENCENT has chosen not to simply chase benchmark rankings, but instead focuses on enhancing model capabilities while maintaining attractive cost efficiency in inference. Management pointed out that as frontier AI models converge in benchmark performance, market demand for models with strong inference economics—particularly DeepSeek and Hunyuan (Hy)—remains stable.

The Hunyuan model is expected to enter a faster iteration cycle, with version HY4 likely to be released within the next three months, further narrowing the capability gap with top domestic models featuring larger parameter scales. Regarding the WeLM underlying model used by WeChat's AI assistant "Xiaowei," management emphasized that it is designed for high cost efficiency, requiring fewer computing resources and compatible with a wider range of GPU types. Therefore, inference costs remain controllable as usage scale expands.

AI Application Rollout: Dual-Track Progression of WorkBuddy and Xiaowei

At the level of AI application commercialization, management positions WorkBuddy as one of TENCENT's most significant AI application opportunities. The product benefits from increased user engagement, improved monetization capabilities, and a positive feedback loop formed with the Hunyuan model ecosystem—data generated from real user interactions helps enhance model intelligence, thereby improving product performance and user experience. Management stated that monetization of WorkBuddy is still in its early stages, but there is significant room for expansion in user base, feature set, and experience optimization.

WeChat's AI assistant Xiaowei is also in its early development stage, with TENCENT adopting a prudent and gradual approach to rolling out features and integrating into the ecosystem. Management highlighted its structural advantages, including lower user acquisition costs and inference costs. Xiaowei will be gradually opened to more users, with full-scale promotion potentially achievable as early as the fourth quarter of 2026, though the specific timeline will also be influenced by partner progress. Potential application scenarios currently being explored include travel planning, price-comparison shopping, and other daily consumer productivity scenarios.

Earnings Forecast Downgrade: Depreciation Pressure Erodes Profit Margins

Aggressive AI investment, while driving long-term growth expectations, has also brought near-term earnings pressure. The report lowered its non-IFRS net profit forecasts for TENCENT from 2026 to 2028 by 2% to 6%, mainly due to a significant acceleration in depreciation expenses over the coming quarters. Analysts currently expect TENCENT's non-IFRS net profit in 2027 to remain flat year-over-year.

Specifically, TENCENT's non-IFRS diluted earnings per share for 2026 and 2027 are now expected to be RMB 29.30 and RMB 29.81, respectively (previously forecast at RMB 29.80 and RMB 31.20). In terms of cloud and enterprise service revenue, TENCENT's cloud revenue growth rate is expected to accelerate further from the second half of 2026 into 2027, surpassing the year-over-year growth rate of slightly above 20% recorded in the second quarter of 2026, driven by increased available computing power and the accelerated rollout of WorkBuddy.

Bank of America Merrill Lynch maintains its Buy rating on TENCENT and a target price of HK$780, based on a sum-of-the-parts valuation method, corresponding to approximately 23 times the forecast P/E ratio for 2026. The analysts listed three near-term catalysts: the upgrade of the Hunyuan model within the next three months, the large-scale promotion of Xiaowei within the next six months, and the accelerated growth of WorkBuddy and cloud revenue within the next twelve months.