Goldman Sachs: Explosion in China AI Token Demand; Top Pick is Cloud & Data Centers

Overview

This report presents Goldman Sachs' latest strategic outlook on the China Internet and AI model industry. The analysis centers on five core debates: US-China AI model competition, domestic price wars, token growth drivers, hyperscaler capex prospects, and consumer-grade AI Agent formats. Based on the view of surging AI inference demand, the report continues to designate 'Cloud & Data Centers' as the top sub-sector pick for H2 2026, believing that the fundamentals or valuations of Alibaba and Tencent are near their bottoms. Meanwhile, considering weak macro consumption, the report has adjusted sub-sector priorities by elevating the ranking of Gaming & Entertainment and adding NetEase, JD.com, and MiniMax as key recommended targets.

Core views

AI Models & Token Economy: The report forecasts that China's daily token consumption will reach 350 trillion by end-2026 (monthly average of 10.5 quadrillion), primarily driven by enterprise-grade Agents. Although Chinese AI models' share of token usage on global third-party platforms like OpenRouter has climbed to nearly 50%, their share of the revenue pool remains in the single digits due to pricing at only 10%-25% of US SOTA models. The report suggests the market is bifurcating into two tiers: high-value scenarios utilizing frontier models, while SMEs and Agent tasks tend to favor cost-effective Chinese models or Flash models. Multimodality (e.g., video generation) is becoming key to enhancing pricing power; for instance, Seedance 2.0 achieved an annualized revenue of USD 1.7 billion in a short period. Sub-sector Allocation Strategy: Cloud & Data Centers maintain first priority, underpinned by rising cloud prices, spreading token demand, and accelerated capex release from hyperscalers in H2 2026 (BBAT full-year capex estimated at USD 100 billion). Given soft macro consumption trends in Q2 and pressure on e-commerce advertising, the report elevates 'Gaming & Entertainment' to second priority, valuing its counter-cyclical characteristics and profit elasticity from declining channel fees; 'E-commerce & Mobility' is downgraded to third, though JD.com's earnings recovery opportunity in H2 2026 is emphasized. The AI Model sector remains fourth, with a focus on recommending MiniMax, bullish on its omni-modal capabilities and ARR growth potential driven by the new M3/Hailuo 3 models. Tech Giant Fundamentals Assessment: For internet giants that have seen stock price corrections since the start of the year, the report believes downside risks have been fully priced in. For Alibaba, despite near-term pressure on CMR, accelerating cloud revenue growth and MaaS business restructuring are expected to drive EPS to bottom out and rebound in H2 2026. For Tencent, although AI investments weigh on near-term profits, its valuation multiple has fallen to a historical low of around 12x; with Hunyuan model iterations, WeChat AI Agent pilots, and increased capex, room for valuation repair opens up. For MiniMax, despite facing price wars and volatility from lock-up expiry, the report maintains that its risk-reward profile remains attractive due to unique architectural efficiency and omni-modal advantages, with the target price implying significant upside.

Analysis framework

The report employs an analytical framework combining 'top-down industry trends + bottom-up individual stock fundamentals.' At the industry level, it quantitatively assesses AI commercialization progress by tracking token invocation data on third-party platforms like OpenRouter, model pricing changes, and ARR growth rates; it judges the cyclical position of infrastructure build-out by comparing capex/revenue ratios and chip supply rhythms between US and Chinese hyperscalers. At the individual stock level, it uses SOTP (Sum-of-the-Parts) valuation to price diversified internet companies and decomposes stock price performance into 'valuation multiple changes' and 'EPS expectation changes' to identify whether declines are due to sentiment-driven overselling or fundamental deterioration, thereby capturing mean reversion or growth inflection opportunities.

Methodology notes

• Valuation MethodologySOTP Valuation

  Sum-of-the-Parts Valuation

  For internet giants with diversified businesses (e.g., e-commerce, cloud, gaming), each business segment is valued separately and then summed. This more accurately reflects the value of new businesses like AI, avoiding it being obscured by low valuations of legacy businesses; it is the core method used in the report to derive target prices for Alibaba, Tencent, and Meituan.

• Company Fundamentals & Financial Framework

  Stock Price Attribution Analysis (Valuation vs. Earnings)

  Decomposes YTD stock price movements into 'P/E multiple changes' and 'EPS expectation changes.' The report uses this to determine that Alibaba's decline is mainly due to EPS downgrades while Tencent's is mainly due to valuation compression, leading to the conclusion that both are near their bottoms.

• Industry/Sector Analysis FrameworkPenetration S-curve

  Exponential Growth in Token Consumption

  Uses the explosive growth in token consumption to gauge the penetration stage of AI applications. The report notes that Agentic AI will drive token demand from linear to exponential growth (reaching 120 quadrillion/month globally by 2030), serving as a leading indicator for judging the sustainability of cloud infrastructure demand.

• Competition & Strategy Framework

  AI Model Five-Dimension Competitiveness Radar Chart

  The report constructs an evaluation framework comprising five dimensions: organizational efficiency, financial strength, model intelligence, cost efficiency, and multimodal capabilities. This is used to horizontally compare the differentiated competitive advantages of players like Alibaba, Tencent, and MiniMax, rather than relying solely on single benchmark scores.

Asset mapping & comparison

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

• Alibaba (BABA/9988.HK)

  Top pick in Cloud & Data Centers, APAC Conviction List

  Strengths

  Full-stack AI player, accelerating cloud revenue growth, MaaS ARR target of USD 4.4bn, EPS downgrade cycle bottoming

  Weaknesses

  CMR dragged by near-term retail slowdown, AI investments widening losses in 'Others' segment

  Comparison

  Compared to Tencent, Alibaba has earlier deployment in enterprise AI and cloud infrastructure with higher earnings visibility

  Risks

  Retail GMV growth missing expectations, cloud revenue growth slowing, poor execution of strategic investments

• Tencent (0700.HK)

  Top pick in Gaming & Entertainment, significant valuation repair potential

  Strengths

  Unique social ecosystem, ads maintaining high double-digit growth, WeChat AI Agent offering differentiated entry point, valuation at historical lows

  Weaknesses

  Fintech affected by weak consumption, AI investments weighing on near-term profits, unclear monetization path for C-end AI

  Comparison

  Compared to Alibaba, Tencent has stronger C-end traffic moats but faces greater catch-up pressure in B-end cloud business

  Risks

  Delays in game license approvals, AI development progress slower than expected, regulatory policy changes

• MiniMax (0100.HK)

  New key target in AI Model sector, favorable risk-reward skew to upside

  Strengths

  Leading omni-modal capabilities, highest API gross margin among peers, catalyst from new M3/Hailuo 3 models, high visibility on ARR growth

  Weaknesses

  Facing intense price wars, July lock-up expiry may cause stock volatility, weaker financial strength compared to giants

  Comparison

  Compared to independent model makers like Zhipu, MiniMax is more differentiated in multimodality and organizational efficiency

  Risks

  Pricing strategy failure, rapid catch-up in model capabilities by big tech, deteriorating financing environment

• JD.com (JD/9618.HK)

  New key target in E-commerce sector, H2 2026 earnings recovery story

  Strengths

  Self-operated retail moat, 2Q26 being the last tough comp quarter, new growth drivers in food delivery/instant retail driven by supply chain

  Weaknesses

  High base in electronics suppressing near-term growth, subsidy competition potentially eroding margins

  Comparison

  Profit recovery certainty is stronger than Pinduoduo; self-operated model shows more resilience in user experience than Alibaba

  Risks

  Intensifying e-commerce competition, slowing GMV growth, retail margin volatility

• NetEase (NTES/9999.HK)

  New key target in Gaming sector, defensive allocation

  Strengths

  Structural profit uplift from lower channel fees, stable evergreen games, new title 'Sea of Remnants' launching in 3Q26

  Weaknesses

  Uncertainty in new game performance, AI cost-saving effects on game development still in early stages

  Comparison

  Lower valuation and greater profit elasticity than Tencent; more stable R&D pipeline than small/mid-sized game studios

  Risks

  Game gross billings missing expectations, tightening industry regulation

Key data

• China Daily Token Consumption Forecast350 TrillionExpected to be reached by end-2026, a 2.5x increase from 140 trillion in March
• BBAT 2026 CapexUSD 100 BillionConcentrated mainly in H2, supporting accelerated data center expansion
• MiniMax M3 API Pricing$0.22/Million TokensPrice after permanent reduction; still carries a premium over DeepSeek/MiMo's $0.18 but offers omni-modal advantages
• Alibaba Cloud Revenue Growth+38% YoYFY26 Q4 data; AI-related revenue achieving triple-digit growth for 11 consecutive quarters
• Tencent 2026E P/E12x-13xAt historical and global peer lows, viewed as a signal of valuation bottom
• Chinese Models' Global Token Share~50%Share on OpenRouter platform; was only low single digits at end-2024

Impact & implications

The report believes the structural explosion in AI inference demand will continuously benefit cloud infrastructure providers over the next 12-24 months, especially those with domestic chip adaptation capabilities and MaaS pricing power. For internet platforms, AI is no longer merely a valuation concept but has entered a period of performance verification: companies that can leverage AI to improve ad ROI or cloud margins will see a Davis Double Play, while pure consumer-facing AI applications may face commercialization bottlenecks if they cannot resolve inference cost issues. Investors should focus on structural opportunities arising from the shift from 'traffic logic' to 'token/compute logic,' rather than betting on an overall consumption recovery.

Risks

• Weak macro consumption recovery leading to sustained pressure on e-commerce and advertising revenue
• Intensifying AI model price wars eroding margins for cloud vendors and model companies
• Domestic chip supply ramp-up falling short of expectations, constraining capex implementation pace
• Escalating geopolitical risks restricting access to high-end compute or overseas business expansion
• High inference costs for consumer-grade AI Agents leading to commercialization failures
• Regulatory policy changes affecting game licenses, data compliance, or the platform economy

What to watch

• Actual quarterly capex execution by major Chinese cloud vendors and domestic chip delivery rhythms
• API invocation volumes and ARR ramp-up data following the launch of MiniMax M3 and Hailuo 3 models
• Inflection point in Alibaba CMR growth and magnitude of cloud business margin expansion
• Progress of Tencent WeChat AI Agent pilots and user retention metrics
• JD.com 2Q26 earnings confirming whether it marks the starting point of profit recovery
• Trends in Chinese model token share and pricing changes on platforms like OpenRouter