Supply-chain dependency graph of the AI compute stack

Name: Supply-chain dependency graph of the AI compute stack: node-by-node taxonomy of 580 supply-chain nodes across 10 layers with 562 typed dependency edges, concentration shares and dominant-player attribution per node, chokepoint/bottleneck classification (9 chokepoints, 112 bottlenecks), and quantitative engineering-economics attributes per node (cost adders, format/size limits, throughput and qualification constraints), each cited to a numbered source list
Creator: Chokepoints

CHOKEPOINTSManage supplier listing

Supply-chain dependency graph of the AI compute stack: node-by-node taxonomy of 580 supply-chain nodes across 10 layers with 562 typed dependency edges, concentration shares and dominant-player attribution per node, chokepoint/bottleneck classification (9 chokepoints, 112 bottlenecks), and quantitative engineering-economics attributes per node (cost adders, format/size limits, throughput and qualification constraints), each cited to a numbered source list

Open buyer room →

Sample

Brickroad customers

Sample this dataset before you buy it.

Your sourcing agent asks Chokepoints and files the sample in your catalog — private to you.

Start 7-day trialNothing is charged until the trial ends.

Coverage

Universe, instruments and categories.

Industry: SemiconductorsTechnology Hardware, Storage & Peripherals
Instruments: equities

Dataset card

Type: not stated
Format: Multimodal: graph/tabular (nodes, edges, concentration percentages) wrapped in long-form cited research text
Volume: 6,500+ companies mapped, 580 supply-chain nodes, 562 dependency edges, 10 layers, plus per-node deep-dive pages of roughly 10-15K characters each with 20-30 numbered source citations apiece; extrapolated corpus of roughly 3-9 million words of cited research text
Users: N/A for a product sense of the word — this is an open research atlas with a newsletter rather than a registered-user service; the audience proxy is the newsletter list, which is not disclosed (Request further research on list size)
History: 14 years
Update frequency: not stated
Growth: Active — the atlas carries 2026-vintage figures (for example ~80-85% accelerator revenue concentration in 2026) and openly solicits incoming supply-chain data, broker notes and teardown photos to extend coverage
Launched: ~2025 (est. — no 'since' copy or archive dates in the fetched pages; the site's content is written around the 2024-2026 AI-capex cycle, which bounds it from below)
Delivery: not stated
Entity mapping: not stated
Sample: not stated
Point-in-time: not stated
Licence: not stated

supplier index read from their site