keystone

Semiconductor ecosystem map

A slow, evidence based map of the global chip industry: who the companies are, how big they are, and above all how they depend on each other. Built to unfold like an onion, one layer at a time.

What this is (and is not)

The source infographic groups companies by the job they do (design, fabricate, package, sell). That is a value chain, not an ownership chart. No company here reports to another. The relationships are commercial: who licenses to whom, who fabricates for whom, who buys from whom. The point of this project is to fill in those edges with real evidence and then see where the true chokepoints are.

Files

data/companies.csv              one row per company (18 col shared schema), all 41 costed
data/relationships.csv          one row per directed relationship (the crown jewel), 48 sourced edges
data/gaps.csv                   what no filing would say, and the documents read to establish it
data/sources.csv                provenance registry
data/financials_timeseries.csv  per company per period, for trends rather than a single snapshot
map.json                        this map's prose, storage key, and value chain layers
AGENT-RUNBOOK.md                exact instructions the scheduled agent follows each run
PROGRESS.log                    append only log of what each burst did; newest entry wins
artifact/                       the built page (generated by shared/build_map.py, never hand edited)
README.md                       this file

The onion (layers of work)

Status is written here in prose, which means it can rot. PROGRESS.log and the data are the truth; if this list disagrees with them, they win.

companies.csv schema

column meaning
company display name
ticker primary stock symbol
exchange where it trades
hq_country headquarters country
roles one or more of: fabless-chip, fabless-system, ip-eda, foundry, equipment, component-supplier, osat, idm, oem (semicolon separated)
primary_segment short plain description of what it mainly sells
revenue_usd_b annual revenue in billions USD
net_income_usd_b annual net income in billions USD
market_cap_usd_b market capitalization in billions USD
rd_spend_usd_b annual research and development spend in billions USD
capex_usd_b annual capital expenditure in billions USD
fiscal_year the fiscal year the figures come from
filing_source url or document the figures came from
last_updated date the row was last filled (YYYY-MM-DD)
chokepoint yes if the whole industry has few or no alternatives to this company
notes free text
lens accounting, politics or economics: which of the three lenses this row belongs to
source_tier 1 to 3, how strong the evidence behind the figures is. 1 is a primary regulatory filing. This is the column “sourced or it does not exist” actually rests on

relationships.csv schema

column meaning
from_company the company that acts
to_company the company acted upon
relationship_type licenses-ip-to, provides-eda-to, fabricates-for, supplies-components-to, sells-equipment-to, is-customer-of, packages-for, invests-in, competes-with
description one plain sentence on the specific relationship
evidence_source url or filing where this was found
evidence_date date of that evidence
confidence high, medium, or low
last_updated date the row was written

Direction matters. “TSMC fabricates-for Nvidia” and “Nvidia is-customer-of TSMC” are the same fact from two sides. Prefer recording the supplier side (fabricates-for, supplies-components-to, sells-equipment-to, licenses-ip-to) so the graph reads as flows of goods and value.

Chokepoint queue (research starts here)

The eight companies the rest of the industry has the fewest alternatives to. All eight are now costed and every one carries sourced edges, which is the bar this map sets for a chokepoint claim. Edge counts as of 2026-07-22:

# company edges why it is a chokepoint
1 Nvidia 17 dominant AI accelerator designer
2 TSMC 10 leading edge foundry the world leans on
3 Intel 10 IDM turning into a foundry
4 Arm 8 instruction set inside almost every phone
5 Samsung 8 spans foundry, memory, components and devices
6 Applied Materials 5 largest chip equipment maker
7 SK hynix 5 key supplier of HBM memory for AI
8 ASML 4 only maker of EUV lithography machines

Counts come from relationships.csv and will drift as edges land. The claim that matters is not the number but that none of these eight is flagged on assertion alone.

Free data sources