The Three-Layer Framework
The simplest way to understand AI investing is through a three-layer architecture — similar to how the internet was built in the 1990s. At the base sits infrastructure (cables, servers, browsers); in the middle sit platforms (Amazon, Google); at the top sit applications (Netflix, Uber, Airbnb). AI follows the same pattern.
Investors who understand which layer they are buying — and where capital is flowing — make more consistent returns. Investors who conflate the three layers take on risks they don't understand.
Layer 1: Infrastructure
What it is
The physical and computational foundation of AI: GPUs and specialised AI chips, data centres and real estate, power generation and cooling systems, networking and fibre, and cloud computing platforms.
Key companies
- NVIDIA (NVDA): Designs the GPUs that power AI training — currently the dominant Layer 1 play.
- Broadcom (AVGO): Custom AI chip (ASIC) designer for hyperscalers.
- TSMC: Manufactures the world's most advanced chips for NVIDIA, Apple and others.
- Vertiv (VRT): Data centre power and cooling infrastructure.
- Vistra Energy (VST) / Constellation Energy: Power generation for AI campuses.
Stage in the cycle
Infrastructure peaks early in the AI buildout cycle. NVIDIA's returns from 2022–2025 reflect Layer 1 capital flowing first. The risk: overbuilding eventually compresses margins.
Layer 2: Foundation Models
What it is
The large language models (LLMs) and other foundation AI systems trained on Layer 1 infrastructure. These models are what most people interact with when they use "AI" — ChatGPT, Claude, Gemini, Grok.
Key players
- OpenAI (private) — ChatGPT, GPT-4o, o3
- Anthropic (private) — Claude
- Google DeepMind (Alphabet / GOOGL) — Gemini
- xAI (private) — Grok
- Meta AI (META) — LLaMA open source models
Most foundation model companies are private. Public market access comes via Alphabet (Google/Gemini), Meta (LLaMA), Microsoft (OpenAI investment), and Amazon (Anthropic investment).
Layer 3: Applications
What it is
Software and services built on top of foundation models. This is where AI meets revenue — productivity tools, code generation, customer service, healthcare diagnostics, legal document processing, and more.
Key companies
- Palantir (PLTR): AI-powered enterprise data analytics and government intelligence
- Salesforce (CRM): Agentforce AI CRM
- ServiceNow (NOW): AI workflow automation
- Microsoft (MSFT): Copilot across Office 365
- Adobe (ADBE): AI creative tools (Firefly)
Stage in the cycle
Application layer returns typically lag infrastructure by 18–36 months. The companies that win this layer will be the generation's equivalents of Google, Amazon and Netflix.
"In every technology cycle, infrastructure wins first. Then platforms. Then applications. The returns compound, but the timing is everything."
Where Is Capital Flowing Now?
As of early 2026, Layer 1 (infrastructure) has already seen its initial explosive gains (NVIDIA up 10x from 2022 lows). Capital is beginning to rotate toward Layer 2 (model providers via their public parents) and Layer 3 (application companies). The key question for 2026–2027: which application companies will demonstrate durable AI revenue — not just AI features?
- AI investing is structured in three layers: Infrastructure → Models → Applications. Capital flows through the stack over time.
- Layer 1 (NVIDIA, Broadcom, power companies) led the early cycle. Infrastructure overcapacity risk is growing.
- Layer 2 (models) is mostly private — public access via Alphabet, Meta, Microsoft, Amazon.
- Layer 3 (applications) is where the next decade's compounders will emerge — Palantir, Salesforce, ServiceNow are early leaders.
- Identifying which layer you are in — and knowing when to rotate — is more important than picking individual names.