SCOI v1 · canonical classification
Every notable AI company, plotted on the stack.
One sortable, citation-grade table. Each row is a company's primary structural position across the 10 layers of The Supply Chain of Intelligence™, with its archetype (fortress, refinery, surface, graveyard…) and a link to the long-form case study where one exists.
99 companies
| Company | Dominant layer | All layers | Archetype | Vertical | Citation |
|---|---|---|---|---|---|
| NextEra | L-1 | L−1 | Defensible stack | infra | - |
| Vistra | L-1 | L−1 | Defensible stack | infra | - |
| TSMC | L-1 | L−1L0 | Defensible stack | infra | - |
| MP Materials | L-1 | L−1 | Refinery | infra | - |
| Bechtel | L-1 | L−1 | Defensible stack | infra | - |
| NVIDIA | L0 | L0 | Defensible stack | infra | - |
| AMD | L0 | L0 | Railroad | infra | - |
| Broadcom | L0 | L0 | Railroad | infra | - |
| CoreWeave | L0 | L0 | Railroad | infra | - |
| Equinix | L0 | L0 | Railroad | infra | - |
| AWS | L0 | L0L4 | Defensible stack | horizontal | - |
| Azure | L0 | L0L4 | Defensible stack | horizontal | - |
| Google Cloud | L0 | L0L2L4 | Defensible stack | horizontal | - |
| Microsoft 365 | L0 | L0L2L4L7 | Defensible stack | horizontal | Case study |
| Apple | L0 | L0L4L7 | Defensible stack | horizontal | - |
| Cloudflare | L0 | L0L4 | Railroad | horizontal | - |
| Bloomberg | L1 | L1L2L3L4 | Defensible stack | finance | Case study |
| Apollo.io | L1 | L1 | Refinery | sales | Case study |
| ZoomInfo | L1 | L1L7 | Refinery | sales | Case study |
| Scale AI | L1 | L1 | Refinery | horizontal | - |
| L1 | L1L7 | Refinery | horizontal | - | |
| Stack Overflow | L1 | L1L7 | Thin stack | code | Case study |
| Getty Images | L1 | L1 | Refinery | creative | - |
| Tempus | L1 | L1L5L8 | Defensible stack | health | - |
| Epic Systems | L1 | L1L4L5 | Defensible stack | health | - |
| Harvey AI | L1 | L1L3L5L8 | Agent (L5+L7) | legal | Case study |
| Google Search | L1 | L1L2L4L7 | Defensible stack | horizontal | Case study |
| GitHub | L1 | L1L4 | Defensible stack | code | - |
| Salesforce | L1 | L1L4L5 | Defensible stack | sales, cx | Case study |
| Sierra | L1 | L1L5L8 | Defensible stack | cx | Case study |
| Klarna | L1 | L1L5L8 | Defensible stack | cx, finance | Case study |
| Decagon | L1 | L1L5L8 | Defensible stack | cx | Racing Sierra for the same F500 CX square, L1 (tenant transcripts) + L5 (CX-stac… |
| Hebbia | L1 | L1L5 | Defensible stack | finance | Finance research workflows on top of internal docs, L1 (firm corpus) + L5 (analy… |
| Rogo | L1 | L1L5 | Defensible stack | finance | Investment-banking analyst copilot. L1 (deal docs) + L5 (pitch/model workflow). |
| Abridge | L1 | L1L3L5 | Defensible stack | health | Clinical documentation. L1 (encounter audio) + L3 (HIPAA / payer-grade) + L5 (EH… |
| Ambience | L1 | L1L3L5 | Defensible stack | health | Same fortress pattern as Abridge, L1+L3+L5 in clinical documentation. |
| Suki | L1 | L1L3L5 | Defensible stack | health | Voice-first clinical AI. L1+L3+L5; contests Abridge / Ambience for EHR-integrate… |
| Clay | L1 | L1L5L8 | Defensible stack | sales | GTM enrichment + workflow. Quietly one of the best-stacked: L1 (data graph) + L5… |
| Mercor | L1 | L1L5 | Agent (L5+L7) | sales | AI recruiting. L1 (candidate corpus) + L5 (hiring workflow); contested by Final … |
| Crosby | L1 | L1L5 | Agent (L5+L7) | legal | AI contract review for ops/commercial. Thin slice of Harvey's board, defensible … |
| Glean | L1 | L1L6L8 | Defensible stack | horizontal | Case study |
| Adobe Firefly | L1 | L1L3L4L7 | Defensible stack | creative | Case study |
| LexisNexis | L1 | L1L4L7 | Defensible stack | legal | Century-old case-law corpus. L1 fortress; Lexis+ AI is the L7 surface on top. |
| Westlaw (Thomson Reuters) | L1 | L1L4L7 | Defensible stack | legal | Westlaw Precision AI sits on the same L1 moat as Lexis, case law + headnotes + K… |
| Bloomberg Law | L1 | L1L4L7 | Defensible stack | legal | Bloomberg's legal arm, dockets, opinions, secondary sources. L1 + L4 distributio… |
| vLex | L1 | L1L7 | Refinery | legal | Global case-law corpus; powers Vincent AI. Refinery racing toward fortress. |
| Fastcase | L1 | L1L7 | Refinery | legal | - |
| Casetext (Thomson) | L1 | L1L5L7 | Defensible stack | legal | Acquired by Thomson Reuters for $650M. CoCounsel is now the L5 surface on Westla… |
| Ironclad | L1 | L1L5 | Defensible stack | legal | CLM incumbent layering AI on top of a per-tenant contract corpus (L1 + L5). |
| LinkSquares | L1 | L1L5 | Defensible stack | legal | CLM + post-signature analytics. Same L1+L5 pattern as Ironclad. |
| Evisort (Workday) | L1 | L1L5 | Defensible stack | legal | Contract intelligence; acquired by Workday, now L1+L5 inside an L4 distribution … |
| DraftWise | L1 | L1L5 | Defensible stack | legal | Drafts off a firm's precedent corpus. L1 (precedents) + L5 (drafting workflow). |
| EvenUp | L1 | L1L5 | Defensible stack | legal | Personal-injury demand letters, L1 (medical/PI corpus) + L5 (case-prep workflow)… |
| Supio | L1 | L1L5 | Agent (L5+L7) | legal | PI case prep, racing EvenUp on the same L1+L5 square. |
| Luminance | L1 | L1L5 | Defensible stack | legal | Diligence + contract analysis; one of the older L1+L5 stacks in legal AI. |
| Kira (Litera) | L1 | L1L5 | Defensible stack | legal | Acquired by Litera, diligence corpus + workflow inside the BigLaw stack. |
| OpenAI | L2 | L2L7 | Defensible stack | horizontal | Case study |
| Anthropic | L2 | L2L3 | Defensible stack | horizontal | Case study |
| Google DeepMind | L2 | L2 | Defensible stack | horizontal | - |
| Meta AI (Llama) | L2 | L2 | Railroad | horizontal | - |
| Mistral | L2 | L2 | Refinery | horizontal | - |
| xAI | L2 | L2 | Refinery | horizontal | - |
| Cohere | L2 | L2 | Refinery | horizontal | - |
| Stability AI | L2 | L2 | Thin stack | creative | Case study |
| Meta | L2 | L2L4L7 | Defensible stack | horizontal | - |
| ChatGPT | L2 | L2L7 | Defensible stack | horizontal | - |
| Claude.ai | L2 | L2L3L7 | Defensible stack | horizontal | - |
| Gemini App | L2 | L2L4L7 | Defensible stack | horizontal | - |
| Copilot | L2 | L2L4L7 | Defensible stack | horizontal | - |
| Midjourney | L2 | L2L7 | Defensible stack | creative | Case study |
| Hippocratic AI | L3 | L3L5 | Defensible stack | health | - |
| Vanta | L3 | L3 | Defensible stack | horizontal | - |
| Drata | L3 | L3 | Refinery | horizontal | - |
| Stripe | L4 | L4 | Railroad | horizontal | - |
| Cursor | L4 | L4L6L8 | Defensible stack | code | Case study |
| Grammarly | L4 | L4L6 | Defensible stack | horizontal | Case study |
| Zapier | L4 | L4L6 | Railroad | horizontal | - |
| Notion AI | L4 | L4L6L8 | Defensible stack | horizontal | - |
| Cresta | L5 | L5L8 | Memory | cx | Real-time agent assist + L8 coaching loop. Defensible as long as the loop stays … |
| Parloa | L5 | L5L8 | Agent (L5+L7) | cx | Voice-first CX agent. L5+L8 stack, racing Sierra/Decagon in EU enterprise. |
| Devin (Cognition) | L5 | L5L7 | Agent (L5+L7) | code | Case study |
| Spellbook | L5 | L5 | Agent (L5+L7) | legal | Contract drafting/review in Word. L5 workflow; thin without L1 firm corpus. |
| Robin AI | L5 | L5 | Agent (L5+L7) | legal | Contract review agent. L5-only; vulnerable to Harvey + CLM incumbents. |
| Eve | L5 | L5 | Agent (L5+L7) | legal | Plaintiff-side litigation copilot. L5; needs to deepen into L1/L8 to survive. |
| Paxton AI | L5 | L5L7 | Agent (L5+L7) | legal | - |
| Leya | L5 | L5 | Agent (L5+L7) | legal | European BigLaw copilot. Harvey-pattern, smaller geography. |
| Definely | L5 | L5 | Agent (L5+L7) | legal | Drafting/definitions tool inside Word. Useful surface, single-layer. |
| CoCounsel (Thomson) | L5 | L5L7 | Defensible stack | legal | Casetext rebrand. Now an L7 surface backed by Westlaw's L1, fortress by acquisit… |
| Lexis+ AI | L5 | L5L7 | Defensible stack | legal | L7 wrapper over LexisNexis's L1. Fortress because the corpus is captive. |
| LangChain | L6 | L6 | Railroad | horizontal | - |
| Profound | L7 | L7 | Surface-only | sales | LLM-recommendation analytics (AEO). Today: Ahrefs-clone L7 dashboard. Master str… |
| Granola | L7 | L7L8 | Memory | horizontal | Meeting notes with L8 personal memory. Memory-orphan risk if L2 (model) and L4 (… |
| Cluely | L7 | L7L8 | Memory | horizontal | Real-time meeting assist. Same L8 memory-orphan exposure as Character.AI, beauti… |
| Perplexity | L7 | L7 | Surface-only | horizontal | Case study |
| Jasper | L7 | L7 | Thin stack | creative | Case study |
| Gamma | L7 | L7 | Thin stack | creative | Case study |
| Chegg | L7 | L7 | Thin stack | edu | Case study |
| Character.AI | L7 | L7L8 | Memory | horizontal | Case study |
| Replika | L7 | L7L8 | Memory | horizontal | - |
Layer assignments reflect a company's primary structural position(s), not every layer they touch. Edits, dissents, and additions welcome, challenge any classification on the Challenge page.
- Diagnose, don't describe
Every classification stands on a per-layer diagnostic test.
Disagree with a row? Use the layer's inclusion / exclusion / removal tests to challenge it.