The AI landscape
Who leads each of the nine layers and how mature it is. There is also an interactive landscape.
| Layer | Key players | Maturity (out of 100) |
|---|---|---|
| Layer 9 · Applications & Interfaces | ChatGPT, Gemini, Microsoft 365 Copilot, Claude, Doubao, Perplexity, Cursor | 70 Mature |
| Layer 8 · Agents | Claude Code, OpenAI Codex, GitHub Copilot coding agent, OpenHands, OpenClaw, Salesforce Agentforce | 40 Growing |
| Layer 7 · Orchestration & Tooling | Model Context Protocol (MCP), LangChain / LangGraph, LlamaIndex, pgvector, Pinecone, Langfuse | 50 Growing |
| Layer 6 · Inference & Optimization | vLLM, SGLang, llama.cpp, Ollama, Amazon Bedrock, OpenRouter, Together AI | 80 Mature |
| Layer 5 · Foundation Models | OpenAI, Anthropic, Google DeepMind, xAI, DeepSeek, Alibaba (Qwen), Mistral AI | 70 Mature |
| Layer 4 · Data | Common Crawl, Hugging Face, Scale AI, Reddit, Wikimedia Enterprise | 60 Growing |
| Layer 3 · Cloud & Infrastructure | AWS, Microsoft Azure, Google Cloud, Oracle Cloud, CoreWeave, Nebius, Kubernetes, Slurm | 80 Mature |
| Layer 2 · Compute Hardware | NVIDIA, AMD, Google (TPU), AWS (Trainium), Huawei (Ascend), SK hynix, Samsung, Micron | 70 Mature |
| Layer 1 · Physical & Energy | TSMC, ASML, Samsung Foundry, Constellation Energy, Equinix, Schneider Electric, Vertiv | 80 Mature |
How these are measured
- Key players
- Organisations and open projects that matter most in each layer, with one line on what each does there. A sourced editorial selection, ordered roughly by importance; not a ranking, and shares appear only where a source gives them.
- Maturity
- A 0–100 score from five yes / partial / no questions about the layer. An editorial assessment using a fixed formula, not a market measurement.
How the maturity score is calculated
score = 100 × (sum of answers) ÷ 5, where yes = 1, partial = 0.5, no = 0
- Standards: Are there established open standards or widely shared formats?
- Choice: Do three or more vendors or projects offer interchangeable options?
- Stable interfaces: Have the core interfaces stayed compatible for 12 months or more?
- Adoption: Is it in broad production use across industries?
- Settled practice: Has the recommended way of doing things been stable for 12 months or more?
Bands: Emerging 0–39 · Growing 40–69 · Mature 70–100. Answers: yes (1), partial (0.5), no (0).
| Layer | Standards | Choice | Stable interfaces | Adoption | Settled practice | Score |
|---|---|---|---|---|---|---|
| L9 Applications & Interfaces | partial | yes | partial | yes | partial | 70 |
| L8 Agents | partial | yes | no | partial | no | 40 |
| L7 Orchestration & Tooling | partial | yes | partial | partial | no | 50 |
| L6 Inference & Optimization | yes | yes | partial | yes | partial | 80 |
| L5 Foundation Models | partial | yes | partial | yes | partial | 70 |
| L4 Data | partial | yes | partial | yes | no | 60 |
| L3 Cloud & Infrastructure | yes | yes | partial | yes | partial | 80 |
| L2 Compute Hardware | partial | partial | yes | yes | partial | 70 |
| L1 Physical & Energy | yes | yes | partial | yes | partial | 80 |
Cost & Economics headline
Big Tech capital spending, last reported quarter: About $202 billion (July to September 2026). Cash that Microsoft, Alphabet, Amazon and Meta reported spending on property and equipment (finance leases not included). Up 108% on the same quarter of 2025. Not all of it is AI, and filings don't split it by layer. Reported spending, not plans: updated automatically from SEC filings. As of 2026-09-30.