AI Infrastructure — Hong Kong Stock Exchange (HKEX) Sector Overview

Date: May 12, 2026

Type: Neutral Landscape | High-Level Overview

Scope: Public companies listed on HKEX with significant exposure to AI compute, cloud, semiconductors, data centers, and AI platforms

Purpose: Internal Research

---

Executive Summary

The Hong Kong Stock Exchange has become the primary listing venue for China's AI infrastructure ecosystem, hosting the country's leading cloud platforms (Alibaba, Tencent, Baidu), its dominant semiconductor foundry (SMIC), and the largest independent data center operators (GDS Holdings, VNET). The sector is experiencing a capex super-cycle catalyzed by China's push for AI self-sufficiency, intensified US export controls, and the "DeepSeek moment" that validated efficient AI could paradoxically increase infrastructure demand.

Key Takeaways:

• China's AI infrastructure TAM is estimated at ~RMB 500-650B (~$70-90B) in 2026E, growing at a ~28-35% CAGR
• Alibaba alone has committed RMB 380B ($52B) over 3 years (2025-2027) to AI infrastructure — more than its total prior cloud capex
• Two pure-play LLM companies — Zhipu AI (2513.HK, $53B mkt cap) and MiniMax (0100.HK, $30B) — listed in Jan 2026, marking the first public LLM stocks globally. Both have rallied 190-600% since IPO
• The sector trades at a 40-60% discount to US AI infrastructure peers on forward P/E, though LLM pure-plays trade at extreme 370-520x revenue multiples
• Domestic chip substitution (Huawei Ascend, SMIC foundry) is the dominant theme, creating both opportunity and structural inefficiency
• Hong Kong-listed AI names have rallied 60-120% from Oct 2024 lows through early 2026 on renewed AI capex commitments

---

1. Market Size & Growth

China AI Infrastructure Total Addressable Market

Sources: Industry estimates based on IDC China, Canalys, company disclosures (Alibaba, Tencent, Baidu capex guides). Market definitions vary; overlap exists between segments.

Key Demand Drivers

1. Hyperscaler AI Capex Surge: China's Big 4 (Alibaba, Tencent, Baidu, ByteDance) have collectively guided to ~RMB 600-800B in AI-related capex over 2025-2027
2. Government "Eastern Data, Western Computing": National initiative to build 8 computing hubs and 10 data center clusters, with RMB 400B+ in planned investment
3. Domestic Chip Substitution: US export controls forcing ~30-40% of AI compute procurement to shift to domestic alternatives (Huawei Ascend, Cambricon, etc.)
4. DeepSeek / Efficient AI Paradox: Jevons paradox — cheaper inference enables new use cases, expanding total compute demand rather than reducing it
5. Sovereign AI Mandates: Government entities and SOEs required to use domestic cloud/AI infrastructure for sensitive workloads

Capex Commitments by Major Players (HKEX-Listed)

---

2. Industry Structure

Value Chain Map — China AI Infrastructure

┌─────────────────────────────────────────────────────────────────────────────┐ │ CHINA AI INFRASTRUCTURE VALUE CHAIN (HKEX) │ ├─────────────────────────────────────────────────────────────────────────────┤ │ │ │ LAYER 1: SILICON LAYER 2: SYSTEMS LAYER 3: FACILITY │ │ ┌──────────────────┐ ┌──────────────────┐ ┌────────────────────┐ │ │ │ Foundry │ │ AI Servers │ │ Data Centers │ │ │ │ SMIC (0981.HK) │ │ Lenovo (0992.HK) │ │ GDS (9698.HK) │ │ │ │ Hua Hong (1347.HK)│ │ (Inspur - A-share│ │ VNET (1235.HK) │ │ │ │ │ │ │ │ SUNeVision (1686.HK│ │ │ │ Chip Design │ │ Networking │ │ │ │ │ │ (Huawei - private) │ │ ZTE (0763.HK) │ │ Power/Cooling │ │ │ │ (Cambricon -A-shr) │ │ │ │ (limited HKEX play)│ │ │ └──────────────────┘ └──────────────────┘ └────────────────────┘ │ │ │ │ LAYER 4: CLOUD / PLATFORM │ │ ┌──────────────────────────────────────────────────────────┐ │ │ │ Alibaba Cloud (9988.HK) | Tencent Cloud (0700.HK) │ │ │ │ Baidu AI Cloud (9888.HK) | Kingsoft Cloud (3896.HK) │ │ │ │ JD Cloud (9618.HK) | China Mobile Cloud (0941.HK) │ │ │ └──────────────────────────────────────────────────────────┘ │ │ │ │ LAYER 5: AI APPLICATIONS / PLATFORMS │ │ ┌──────────────────────────────────────────────────────────┐ │ │ │ SenseTime (0020.HK) | Kuaishou (1024.HK) | Xiaomi (1810) │ │ │ └──────────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────────────────────┘

Where Value Accrues in China vs. US

Key insight: Unlike the US where value concentrates heavily at the chip design layer (NVIDIA captures 70%+ GM), China's value chain is more evenly distributed but with lower overall margins. The domestic chip ecosystem is less mature, reducing chip design moat premiums but also creating supply bottlenecks that benefit foundries and server assemblers.

Industry Concentration

• Cloud IaaS: Top 4 (Alibaba ~30%, Huawei ~19%, Tencent ~16%, China Telecom ~14%) hold ~79% share (Canalys 2024)
• AI Chips: Huawei Ascend dominates domestic (~60%+ of domestic AI chip revenue), with NVIDIA still holding ~50-60% of total China AI chip installed base (legacy)
• Data Centers: GDS + Chindata + VNET collectively hold ~20-25% of third-party wholesale market
• AI Servers: Inspur (~30%), Lenovo (~20%), H3C/New H3C (~18%), Sugon (~12%)

---

3. Competitive Landscape

Top HKEX-Listed AI Infrastructure Companies

Note: Market caps and revenues are approximate as of early May 2026. Revenue figures use most recent reported full-year or TTM.

Company Profiles

Alibaba Group (9988.HK) — The Anchor

• Position: China's #1 cloud provider (~30% IaaS share); largest single source of AI infrastructure capex in China
• AI Strategy: Committed RMB 380B over 2025-2027 to AI infra — exceeding its total prior 10-year cloud investment. Building massive GPU clusters for both internal (Tongyi Qianwen models) and external inference services
• Moat: Ecosystem lock-in (e-commerce + fintech + cloud); largest enterprise customer base in China; proprietary chip design (Yitian 710 server CPU, Hanguang AI chip)
• Recent: Cloud revenue re-accelerated to ~12% YoY growth in FY2025 after years of single-digit growth; AI-related cloud revenue growing 100%+ (off small base)
• Valuation: ~12-14x forward P/E (vs. Amazon ~50x); deep discount reflects Ant Group overhang and geopolitical risk

SMIC (0981.HK) — The Foundry Backbone

• Position: China's largest and most advanced semiconductor foundry; sole domestic option for advanced-node chip production
• AI Relevance: Fabricates Huawei's Ascend and Kirin chips (reportedly at 7nm-equivalent using multi-patterning DUV); critical bottleneck for China's AI chip self-sufficiency
• Moat: Regulatory capture — only domestic foundry with advanced capabilities; US entity-list status ironically creates national champion dynamic
• Recent: Revenue grew ~27% in 2024 on surging domestic AI chip demand; capacity utilization above 85%; expanding 28nm and 14nm capacity aggressively
• Risk: Technology gap vs. TSMC (stuck at 7nm DUV equivalent while TSMC at 2nm EUV); US sanctions could restrict further equipment access
• Valuation: ~20-25x forward P/E; premium to global foundry peers on scarcity value

Lenovo Group (0992.HK) — The Server Play

• Position: World's #3 server vendor; #1 PC vendor; Infrastructure Solutions Group (ISG) is the AI-exposed segment
• AI Relevance: ISG segment manufactures AI servers using both NVIDIA and Huawei Ascend GPUs for Chinese enterprise customers; AI server orders reportedly doubled in FY2024/25
• Moat: Global scale + dual supply chain (can source both Western and Chinese chips); strong enterprise distribution
• Recent: FY2024/25 revenue ~$62B (+20% YoY); ISG revenue grew ~60% driven by AI server demand; guided for continued acceleration
• Valuation: ~11-13x forward P/E; cheapest entry point to AI server theme globally

GDS Holdings (9698.HK) — The Data Center Pure-Play

• Position: China's largest independent (carrier-neutral) data center operator; ~500,000 sqm committed capacity
• AI Relevance: Building "AI-ready" high-power-density facilities (30-50kW per rack vs. traditional 8-10kW); key supplier to hyperscalers (Alibaba, ByteDance, JD are major customers)
• Moat: Land + power access in Tier 1 cities; long-term contracts (8-10 year terms); first-mover in high-density AI facilities
• Recent: International expansion into SE Asia (Malaysia, Indonesia) for AI compute; backlog growing 30%+ YoY
• Risk: High leverage (net debt/EBITDA ~5-6x); execution risk on international buildout
• Valuation: ~15-18x EV/EBITDA; discount to US peers (Equinix ~25x) but premium to historical range

SenseTime (0020.HK) — The AI Platform

• Position: China's leading AI platform company; operates SenseCore (one of China's largest commercial GPU clusters, ~45,000 GPUs)
• AI Relevance: Sells "AI infrastructure as a service" — GPU compute + AI development platform + large models (SenseNova). Revenue shifted from CV/smart city to generative AI infrastructure
• Moat: Installed GPU base; AI development ecosystem; government + enterprise relationships
• Recent: Generative AI revenue now >60% of total (was <10% in 2022); turning toward profitability in 2025-2026E after years of losses
• Risk: Cash burn history; dependence on government contracts; US entity list since 2021
• Valuation: ~10-15x EV/Revenue; loss-making so P/E not applicable; narrative-driven

Newly Listed LLM / Foundation Model Companies (Jan 2026 IPOs)

The most significant development in HKEX's AI ecosystem was the back-to-back listing of China's two leading independent LLM companies in January 2026. These represent the first pure-play large language model companies to go public anywhere globally.

Zhipu AI / Knowledge Atlas Technology (2513.HK) — China's OpenAI

• Business: Founded from Tsinghua University research; develops the GLM model series (ChatGLM, GLM-4, GLM-5.1). Products include z.ai assistant, AutoGLM (AI agent), AMiner (academic search), Zread.ai
• Positioning: Considered China's closest equivalent to OpenAI — strong in both research and commercialization; enterprise-focused with government and SOE customers
• Key Events: GLM-5.1 open-source release (Apr 8, 2026) drove shares +11.5%; Feb 2026 shares dropped ~23% on compute resource shortages and user service complaints
• Backing: Tsinghua University-linked; $400M funding round May 2024 at ~$3B pre-IPO valuation (now trading at ~17x that valuation)
• Bull case: Best-positioned domestic LLM for enterprise/government adoption; strong research team; open-source strategy drives ecosystem lock-in
• Bear case: 524x revenue multiple is extreme; deeply loss-making; intense competition from Alibaba Tongyi, Baidu Ernie, ByteDance Doubao (all free/subsidized)

MiniMax Group (0100.HK) — The Consumer AI Play

• Business: Founded by ex-SenseTime researchers; known for Talkie (AI character chat app, ~11M MAU) and Hailuo AI (text-to-video/music generation). Develops MiniMax-01 series LLMs
• Positioning: Consumer-first AI company — differentiated from enterprise-focused peers. Strongest in multimodal (video, audio, character interaction)
• Key Events: Faced Disney/Universal/Warner Bros. copyright lawsuit (Sep 2025); Anthropic accused MiniMax of data harvesting via fraudulent accounts (Feb 2026); first earnings released Apr 2026
• Backing: Alibaba led $600M round (Mar 2024) at $2.5B pre-IPO valuation; Tencent is also a backer
• Bull case: Leading position in AI-generated video/entertainment; massive consumer engagement; potential ByteDance/TikTok competitor in AI content
• Bear case: 378x revenue with HK$14.6B losses; IP litigation risk; only 415 employees creates key-person risk; consumer AI monetization unproven

Newly Listed AI Hardware / Chip Companies (Jan-Apr 2026 IPOs)

Biren Technology (6082.HK) — China's GPU Challenger

• Business: Shanghai-based fabless GPU designer, founded 2019 by former NVIDIA and Alibaba engineers. Designs BR100 and BR104 GPUs for AI training and high-performance computing
• Positioning: One of only 3-4 credible domestic NVIDIA alternatives (alongside Huawei Ascend, Cambricon, Enflame). BR100 targets data center AI training; BR104 for inference workloads
• Key Events: Added to US Entity List (Oct 2023), cutting off TSMC foundry access; now dependent on domestic foundry (SMIC) for production, constraining performance ceiling
• Backing: State-backed investors; significant government procurement contracts for national AI compute centers
• Bull case: Pure-play bet on China's domestic GPU self-sufficiency; government mandates could force adoption regardless of performance gap
• Bear case: Enormous losses vs. tiny revenue; BR100 performance still lags NVIDIA H100 significantly; SMIC foundry constraints limit next-gen chip potential; entity list blocks international growth

Lightelligence (1879.HK) — Photonic AI Computing

• Business: MIT spinout (founded 2017 by Shen Yichen); develops optical/photonic computing chips that use light instead of electrons for AI inference, promising dramatically lower power consumption
• Positioning: Next-generation AI compute paradigm; if photonic computing delivers on promises, it could leapfrog traditional semiconductor limitations
• Backing: Tencent, Sequoia Capital
• Bull case: Paradigm-shifting technology; massive power efficiency advantage could make it critical for inference-at-scale; very early stage with exponential upside
• Bear case: Pre-commercial technology; RMB 106M revenue is negligible; photonic computing has been "5 years away" for decades; extreme valuation risk

---

LLM IPO Pipeline (Not Yet Listed)

Valuation Context: LLM Stocks vs. Global Peers

The HKEX-listed LLM companies trade at extreme revenue multiples relative to even the most highly-valued US AI peers. This reflects: (1) very early-stage revenue bases, (2) speculative retail flows via Southbound Connect, (3) scarcity premium as the only publicly-traded pure-play LLM stocks globally, and (4) expectation of rapid revenue ramp as AI adoption accelerates in China.

Key risk: These valuations require explosive revenue growth (10x+ over 2-3 years) to normalize multiples. If monetization disappoints or competition from subsidized big-tech AI (Alibaba, ByteDance, Tencent all offer free/low-cost LLM access) compresses pricing power, these stocks face significant de-rating risk.

---

4. Key Themes & Trends

Theme 1: The DeepSeek Inflection (Jan 2025 onward)

DeepSeek's V3 and R1 models demonstrated that competitive AI performance was achievable at 10-50x lower training costs than Western models. Initial market reaction was to sell AI infrastructure stocks (Jevons paradox misunderstood). However:

• Actual impact: Cheaper inference unlocks mass-market AI adoption in China, dramatically expanding total compute demand
• Beneficiaries: All HKEX cloud/infra names — Alibaba, Tencent, Baidu saw AI cloud inquiries surge 3-5x post-DeepSeek
• Net result: Chinese tech stocks rallied 40-80% from Jan-May 2025 as market re-rated AI adoption potential

Theme 2: US Export Controls — Constraint Creates Opportunity

• Timeline: Oct 2022 (initial), Oct 2023 (expanded), Jan 2025 (tightened) — progressively restricting NVIDIA A100/H100/H200/B200 exports to China
• Impact on HKEX: Forces domestic substitution to Huawei Ascend → benefits SMIC (foundry), Lenovo (dual-source servers), domestic chip designers
• Workarounds: Some cloud providers still access high-end NVIDIA chips via Southeast Asian subsidiaries; Huawei Ascend 910C reportedly at 70-80% of H100 performance for training
• Investment implication: Long-term structural tailwind for SMIC, Huawei ecosystem, and any HKEX company building Ascend-compatible software stacks

Theme 3: "Eastern Data, Western Computing" (东数西算)

• National policy directing data center construction to western provinces (Guizhou, Gansu, Inner Mongolia) where power is cheap
• RMB 400B+ in planned infrastructure investment across 8 national computing hubs
• HKEX beneficiaries: GDS Holdings, VNET, China Mobile (building massive western facilities), Lenovo (server procurement)

Theme 4: Hong Kong as an AI Capital Market Hub

• Hong Kong has become the definitive listing venue for China's AI ecosystem
• HKEX relaxed listing rules for pre-revenue tech companies (Chapter 18C) — directly enabling AI model company IPOs
• Jan 2026 milestone: Zhipu AI (2513.HK) and MiniMax (0100.HK) listed on consecutive days (Jan 8-9, 2026), marking the first pure-play LLM companies to go public globally
• IPO pipeline: Moonshot AI (Kimi) and StepFun reported to be exploring HKEX listings (Bloomberg, Feb-Mar 2026); DeepSeek remains private ($10B valuation)
• Connect programs allow mainland China investors to buy HK-listed AI stocks via Southbound Connect
• Both Zhipu and MiniMax have seen extraordinary post-IPO rallies (594% and 191% respectively), validating HKEX as the AI capital market

Theme 5: The Inference Buildout (2025-2027 Focus)

• China's AI market is pivoting from training-focused to inference-focused infrastructure
• Inference is more latency-sensitive, distributed, and heterogeneous — favoring edge compute (China Mobile, Xiaomi) and cloud platforms (Alibaba, Tencent)
• Lower per-unit compute cost of inference enables mass deployment but requires 5-10x more total GPU-hours as AI agents proliferate

---

5. Valuation Context

HKEX AI Infrastructure vs. US Peers

Why the Discount Persists

1. Geopolitical risk premium: US-China tensions, potential sanctions escalation
2. Export control overhang: Uncertainty over future chip access limits capex visibility
3. Governance concerns: VIE structures, state influence on capital allocation
4. Lower structural margins: More competitive market, price wars in cloud
5. Growth deceleration risk: China macro slowdown could compress enterprise IT budgets
6. Liquidity discount: Lower free float, less institutional ownership vs. US large-caps

Historical Valuation Range (Hang Seng Tech Index)

• Current: ~18x forward P/E (May 2026)
• Trough: ~10x (Oct 2022, regulatory + macro bottom)
• Peak: ~45x (Feb 2021, pre-crackdown)
• 5-year average: ~22x

Recent M&A Transaction Multiples

• Chindata privatization (2024): ~15x EV/EBITDA
• ChinaCache acquisition (2024): ~8x EV/EBITDA
• GDS International spin/stake sale (2025): implied ~18x EV/EBITDA for international DC assets

---

6. Risks

---

7. Investment Implications

Best Risk/Reward Opportunities

1. Alibaba (9988.HK): Trading at ~12x forward P/E with the most aggressive AI capex commitment in China. Cloud re-acceleration + AI revenue inflection = multiple expansion catalyst. Key risk: capital discipline concerns given massive capex.

1. SMIC (0981.HK): Structural beneficiary of domestic chip substitution. Scarcity premium justified given no alternative domestic advanced foundry exists. Key risk: technology plateau at 7nm if equipment access further restricted.

1. Lenovo (0992.HK): Cheapest global entry point to AI server theme at ~11x P/E. ISG growth accelerating and benefits from both NVIDIA and Huawei ecosystems. Key risk: thin server margins; revenue lumpy.

1. GDS Holdings (9698.HK): Pure-play on China data center buildout + international expansion. De-leveraging story with improving unit economics. Key risk: high leverage; interest rate sensitivity.

Key Debates

Catalysts to Watch

• Q2-Q3 2026: Alibaba, Tencent, Baidu cloud revenue disclosures (AI mix breakout)
• H2 2026: Moonshot AI (Kimi) and StepFun HKEX IPOs — expanding the LLM pure-play universe
• H2 2026: Next-gen Huawei Ascend chip (910D) launch — performance vs. NVIDIA Blackwell
• H2 2026: Zhipu AI and MiniMax first full-year post-IPO earnings — revenue trajectory will validate or deflate multiples
• Ongoing: US export control revisions under current administration
• H1 2027: SMIC N+2 node (5nm-equivalent) production viability
• 2027: Potential DeepSeek IPO on HKEX (transformative for sector narrative; currently private at $10B)

---

Appendix: Sector Comparison — HKEX vs. US AI Infrastructure

---

Disclaimer: This overview is for internal research purposes. Financial data is approximate and based on publicly available information as of May 2026. Verify specific figures against company filings before use in external materials. Market caps and prices subject to daily fluctuation.