[Executive Summary]

1. Paradigm Shift: From Scale-Up to Realized ROI

• Transition to the Monetization Phase: The AI investment cycle has evolved beyond Phase 1, which was dominated by raw parameter scaling and compute capacity expansion. It is now transitioning into Phase 2, characterized by resolving physical infrastructure bottlenecks, optimizing inference unit economics, and driving measurable ROI in enterprise production environments.
  
• The Pivot: The primary market imperative has shifted from “Who can build the largest model?” to “Who can structurally lower inference costs and generate durable cash flow and productivity gains?”

2. Five Core Layers of the AI Value Chain: Market Dynamics & Investment Directives

① Layer 1. Semiconductor & Silicon Architecture

• Core Value Chain: GPUs, HBM, Custom ASICs, Advanced Packaging, Foundry
  
• Key Players: NVIDIA, AMD, TSMC, Samsung Electronics, SK Hynix, Micron Technology, Broadcom, Marvell Technology, Arm
  
• Key Trend: Diversification away from pure general-purpose GPU dominance (NVIDIA) toward High Bandwidth Memory (HBM4), hyperscaler Custom ASICs (Broadcom, Marvell), advanced packaging (TSMC CoWoS), ultra-high-speed optical interconnects, and ultra-low-power edge silicon.
  
• Key Risks: Massive CAPEX Cycles, Geopolitical/Supply Shocks
  
• Investment Focus: Anchor allocations in large-cap equities (Public/PE) possessing proprietary moats across advanced packaging, foundries, and the HBM complex; allocate venture capital to high-throughput interconnect, chiplet, and energy-efficient NPU startups.

② Layer 2. Compute Infrastructure & Energy

• Core Value Chain: Hyperscale Cloud, Neoclouds, Power Grids, Liquid Cooling, SMRs
  
• Key Players: Microsoft Azure, AWS, Google Cloud, CoreWeave, Equinix, Vertiv, Schneider Electric, Eaton, GE Vernova, Constellation Energy, Doosan Enerbility, HD Hyundai Electric, LS ELECTRIC
  
• Key Trend: The transition toward gigawatt-scale AI data centers has elevated the physical stack—grid interconnects, baseload nuclear/SMR capacity, direct-to-chip liquid cooling, and power distribution systems—into the primary bottleneck. Concurrent rise and consolidation of AI-specialized Neocloud platforms (e.g., CoreWeave).
  
• Key Risks: Protracted Interconnect Timelines, Regulatory Hurdles, Overbuilding
  
• Investment Focus: Build core exposure via public equities and real asset strategies across grid equipment, thermal management, and baseload generation; deploy venture capital into AI-native power management software and next-generation thermodynamic cooling solutions.

③ Layer 3. Foundation Models & Core Engines

• Core Value Chain: Frontier LLMs, Open-Weight Architectures, SLMs
  
• Key Players: OpenAI, Anthropic, Google (Gemini), Meta (Llama), xAI (Grok), Alibaba (Qwen), DeepSeek
  
• Key Trend: Bifurcation between capital-intensive Frontier AI labs and cost-disruptive Open-Weight ecosystems (Meta, DeepSeek). R&D expenditure has shifted from pre-training parameter expansion toward test-time compute (reasoning), native multi-modal orchestration, and post-training inference optimization.
  
• Key Risks: Rapid Commoditization, High Inference Cost Burdens
  
• Investment Focus: Restrict direct model risk to hyperscalers and mega-cap platforms with self-funding balance sheets and deep distribution moats; selectively allocate late-stage growth capital to startups focused on domain-specific fine-tuning, model distillation, and high-efficiency inference infrastructure.

④ Layer 4. Enterprise AI & Agentic Infrastructure

• Core Value Chain: Multi-Agent Orchestration, RAG, Data Governance, Security
  
• Key Players: Microsoft, Databricks, Snowflake, Salesforce, ServiceNow, Scale AI, LangChain
  
• Key Trend: Evolution from single-turn conversational interfaces to autonomous multi-agent orchestrations. Convergence of enterprise data estates, RAG frameworks, deterministic guardrails, AI governance, and agent ops into a unified software stack.
  
• Key Risks: Open-Source Alternatives, Hyperscaler Platform Bundling
  
• Investment Focus: Concentrate growth capital (Series A–C) on orchestration layers, enterprise-grade AI security, and LLMOps solutions deeply embedded within core enterprise systems, creating defensible, high-switching-cost flywheels.

⑤ Layer 5. Vertical Applications & Physical AI

• Core Value Chain: Domain-Specific Solutions, Autonomous Mobility, Humanoid Robotics
  
• Key Players: Palantir, Cursor (Anysphere), Harvey, Tempus AI, Tesla (Optimus), Waymo, Figure AI, Boston Dynamics, Agility Robotics, Unitree
  
• Key Trend: Market consolidation around defensible vertical platforms (e.g., software engineering, legal discovery, computational biology) that replace “thin wrappers” with deep domain workflows. Simultaneous commercial deployment of Vision-Language-Action (VLA) and embodied AI models across autonomous mobility and robotic automation.
  
• Key Risks: “Thin-Wrapper” Vulnerability, Production Unit Economics
  
• Investment Focus: Target vertical platforms anchored by proprietary transactional data, real-world robotic control stacks, and embodied AI systems; and leading traditional-industry companies (Public) whose profit margins are structurally improving through AI transformation (AX).

3. Investment Implications & Strategic Imperatives

• Value Migration Across the Stack: Alpha is shifting from raw GPU scarcity toward the critical Enablers that unlock physical capacity (Layer 2) and the enterprise Monetizers that generate tangible cash flows (Layers 4 & 5).
  
• The Physical Bottleneck as a Structural Supercycle: Power generation, grid transmission, substations, and industrial cooling are no longer adjacent sub-sectors; they are direct, mission-critical gating factors for the broader AI CAPEX buildout.
  
• Workflows Over Weights: Standalone model intelligence is increasingly commoditized. Enterprise value accrues to platforms that embed directly into system-of-record operational workflows, establishing enduring data flywheels and high structural switching costs.
  
• Physical AI Deployment Thresholds: Underwrite physical and robotic automation against unit economics, industrial reliability, and series-production capacity, rather than controlled-environment proof-of-concept demos.
  
• Institutional Barbell Asset Allocation
  – Core Portfolio (Public / PE): High-conviction anchor positions in Tier-1 semiconductor monopolies and foundational utility, grid infrastructure, and thermal management providers with durable free cash flow (FCF) yields.
  – Alpha Portfolio (Private / VC): Highly selective underwriting of Series A–C companies operating at normalized valuations with proven ARR growth across autonomous enterprise orchestration, mission-critical AI security, and production-ready Physical AI (embodied robotics).

* Sources: NVIDIA, TrendForce, Morgan Stanley, Sequoia Capital Research, etc.