01 · AI MARKET ATLAS
The AI MarketStack
Six connected layers, seventy-two mapped companies and an evidence layer around capital, compute, model competition and adoption. The market data adds context; the company map remains the core of the atlas.
6 layers72 companies79 company refs7 market sources
02 · MARKET EVIDENCE
Put hard market numbers beside the company map.
External market evidence gives scale to the six-layer map. These numbers are measured or reported observations and are intentionally separated from the atlas's editorial 1–5 scores.
Private AI investment · 2025USD billions
Stanford HAI ↗United States$285.9B
Hyperscaler compute concentrationTop-five combined share · 2024–2025
Epoch AI ↗03 · VALUE CHAIN PROFILE
Each layer has a different strategic shape.
Every cell is the average editorial score across the twelve companies in that layer. Use the matrix to compare shape — not market share, revenue or investment performance.
Editorial assessmentDirectional 1–5 lens
15
Higher color intensity = higher average score within that lens.
CapabilityEcosystemEnterpriseOpennessEconomicsDistributionEvidence
Compute12 companies
2.0avg refs / company
Data12 companies
2.0avg refs / company
Models12 companies
2.8avg refs / company
AI Infrastructure12 companies
2.0avg refs / company
Applications12 companies
2.0avg refs / company
Distribution & Services12 companies
2.0avg refs / company
04 · LEADERS BY LAYER
Show the names, role and evidence depth together.
Four visible leaders per layer make the map useful before a company drawer is opened. Products, role, capability and connected reference count remain visible on the surface.
01Compute
12 companies02Data
12 companies03Models
12 companies04AI Infrastructure
12 companies05Applications
12 companies06Distribution & Services
12 companies05 · OPENNESS
Openness changes by layer — and by competitive context.
The stacked bars are derived directly from the seventy-two company profiles. External frontier-model evidence adds context without replacing the company-level openness labels.
Stanford HAI ↗OpenMixedClosed
210
651
723
831
48
12
06 · PROFILE ANALYTICS
Use more of the data already inside the 72 company profiles.
These views are derived from existing profile fields. Keyword-based buyer, deployment and business-model categories are multi-label, so their totals can exceed seventy-two.
Usage / consumption24
Subscription / SaaS23
Services / consulting17
Hardware4
Licensing / platform10
Freemium4
2.9 avg products
2.0 avg references
4.4 enterprise score
2.8 avg products
2.0 avg references
4.3 enterprise score
2.8 avg products
2.8 avg references
4.3 enterprise score
2.5 avg products
2.0 avg references
4.3 enterprise score
2.2 avg products
2.0 avg references
4.7 enterprise score
2.8 avg products
2.0 avg references
5.0 enterprise score
07 · FRONTIER MODELS
Positioning map + independent performance snapshot.
The scatter is our editorial positioning map; the adjacent Arena Elo ranking is external evidence. Keeping them side by side makes the distinction visible rather than burying it in footnotes.
Model-company positioningEditorial map · bubble size = distribution
↗ more open↑ more capable
High capability
More closed
More closed
High capability
More open
More open
Lower capability
More closed
More closed
Lower capability
More open
More open
Arena EloMarch 2026 · independent snapshot
Stanford HAI ↗01Anthropic1503
02xAI1495
03Google1494
04OpenAI1481
05Alibaba1449
06DeepSeek1424
3.3%closed-vs-open top-model gap6 of the top 10 Arena models were closed.
08 · DISTRIBUTION + ADOPTION
Distribution strength matters more as adoption becomes mainstream.
The ranking is derived from the company dataset. External adoption signals explain why distribution and enterprise channels matter increasingly across all six layers.
Stanford HAI ↗013 refs
023 refs
033 refs
042 refs
052 refs
062 refs
072 refs
082 refs
092 refs
102 refs
112 refs
122 refs
09 · COMPUTE CONTEXT
Physical constraints make Compute structurally different.
Capital intensity, build time and hardware scaling explain why the first layer behaves differently from software layers above it.
Epoch AI ↗10 · EVIDENCE ARCHITECTURE
Reference depth is metadata — and now it is visible as metadata.
Reference count is not a quality score. The atlas exposes both profile depth and source-type composition so users can see what kind of evidence sits behind the surface.
11 · R&D CONTEXT
Industry leads frontier model production, while geography still matters.
These market-level signals contextualize the Models and Infrastructure layers without replacing company-level records.
Stanford HAI ↗12 · LAYER EXPLORER
Go from market structure to the full company record.
Six profiles are shown first to keep the report readable. Search, filter or expand the selected layer when you need the complete twelve-company view.
Open-weight frontier labDeepSeekOpen model + API provider
5.0Capability
Open-weight frontier lab creating strong capability and price pressure.
- Core products
- DeepSeek V4 · DeepSeek API
- Primary buyer
- Developers · researchers · enterprise
- Business model
- API + open weights
- Deployment
- Managed API · self-hosted
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Frontier + multimodal modelsAlibaba QwenQwen model family
5.0Capability
Broad multimodal family paired with Alibaba Cloud distribution.
- Core products
- Qwen · Qwen-VL · Qwen Image
- Primary buyer
- Developers · enterprise · consumers
- Business model
- Cloud API + open models
- Deployment
- Alibaba Cloud · open deployments
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Frontier model labAnthropicClaude model family
5.0Capability
Frontier provider strongly positioned in coding, knowledge work and enterprise deployment.
- Core products
- Claude · Claude API · Claude Code
- Primary buyer
- Developers · enterprise · knowledge workers
- Business model
- API + subscriptions + enterprise
- Deployment
- Managed cloud + cloud partners
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Frontier model labGoogle DeepMindGemini models + research
5.0Capability
Frontier model organization integrated with Google's cloud, search, productivity and device distribution.
- Core products
- Gemini · Gemini API
- Primary buyer
- Developers · consumers · enterprise
- Business model
- Cloud/API + product ecosystem
- Deployment
- Managed cloud · edge variants
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Open-weight ecosystemMetaLlama family
5.0Capability
Open-weight strategy designed for broad developer, cloud and device distribution.
- Core products
- Llama
- Primary buyer
- Developers · researchers · enterprise
- Business model
- Strategic ecosystem / licensing
- Deployment
- Self-hosted · cloud partners · edge
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Frontier + open modelsMistral AIEuropean model platform
5.0Capability
European model provider combining frontier, specialist and open deployment options.
- Core products
- Mistral models · Codestral · OCR
- Primary buyer
- Developers · enterprise · governments
- Business model
- API + enterprise + open models
- Deployment
- Cloud · private · edge
Capability
Ecosystem
Enterprise
Openness
Economics
Distribution
Models · sorted by Capability
13 · METHOD + SOURCES
Facts, external evidence and editorial assessment stay separate.
Company facts and references remain in the original dataset. External market evidence provides scale and context. Directional scores remain explicitly labeled as editorial assessment.
Products, buyers, deployment and company-specific claims linked to 79 references.
Capital, compute, adoption and frontier-performance data linked to named external research.
Directional 1–5 lenses for comparison. Not market share, revenue or investment advice.
Exact counts where possible; keyword-normalized multi-label charts explicitly identified as normalization.
01
Stanford HAI2026 AI Index Report
Checked 31 Aug 2026
↗02Stanford HAI2026 AI Index — Economy
Checked 31 Aug 2026
↗03Stanford HAI2026 AI Index — Technical Performance
Checked 31 Aug 2026
↗04Stanford HAI2026 AI Index — Research and Development
Checked 31 Aug 2026
↗05Stanford HAI2026 AI Index — Responsible AI
Checked 31 Aug 2026
↗06Epoch AITrends in Artificial Intelligence
Checked 31 Aug 2026
↗07Epoch AIFive hyperscalers now own over two-thirds of global AI compute
Checked 31 Aug 2026
↗