43 — The Cross-Industry Influence Map
Which sectors drive change in which others · 25 sectors · research date 2026-09-15
0. Method, and a negative finding that has to be stated first
The schema does not yet carry this information
The obvious way to build an influence map is from the related_trends foreign keys. That
does not work, and the reason is a Phase 1 finding rather than an inconvenience.
Across all 500 records, related_trends contains zero cross-sector links. Every one of
the 1,535 references points to a trend inside the same industry (a single reference is
unresolved). Twenty-five sector analysts each built a dense intra-sector graph and none
built an inter-sector edge, because nothing in the record schema or the research contract
asked them to. The consequence is that the database knows about the AI→power→real-estate
chain — it is described in prose in at least nine records — but cannot traverse it.
Phase 2 requirement: add a typed cross_sector_links field with an explicit relation
(constrains, funds, supplies, regulates, substitutes_for, measures) and require
it at QA. Until then, every map of this kind must be reconstructed from text, which is what
follows.
The three evidence sources actually used
| Source | What it gives | Volume |
|---|---|---|
| Explicit cross-sector references in trend prose | An analyst in sector X naming sector Y as cause or constraint | 15 directed references across 12 sector pairs — e.g. 18→05 (T-18-11, T-18-20), 25→09 (T-25-02, T-25-06), 21→07 (T-21-01, T-21-06), 09→01 (T-09-19), 20→05 (T-20-19) |
| Shared entities across sectors | Structural coupling: the same firm or regulator appearing in two sectors' records | 51 of 994 entity records sit in more than one industry; 42 named companies appear in the trend records of three or more sectors |
| Documented causal chains in dossier §8 and §10 | The mechanism, with dates and figures | The primary source for every chain in §2 |
Shared-entity coupling by itself is a weak signal — Google appears in thirteen sectors because it is large, not because those sectors are coupled. It is used here only to corroborate a mechanism established elsewhere. The strongest entity-derived pairs are: 01-03 (12 shared companies), 14-16 (12), 05-18 (10), 15-17 (10), 09-18 (10), 07-11 (10), 09-13 (9), 07-21 (9), 01-11 (9).
1. Sector 11 is a capital layer and must never be summed with the verticals
This is the most important structural caveat in the entire research programme and it is stated before anything else because the error it prevents is both easy and expensive.
Sector 11 (Startups, venture capital & private markets) does not measure an industry. It measures the financing of other industries. Its headline metrics are denominated in the same dollars that appear again inside the verticals:
- T-11-02: OpenAI and Anthropic alone took $217bn, 43% of all H1 2026 global venture funding. Those companies and that capital are also the subject of sector 01's records.
- >70% of Q2 2026 venture capital went to AI-focused companies. Sector 13's humanoid funding, sector 04's $300m+ of autonomous-SOC rounds, sector 20's climate-tech totals and sector 18's proptech figures are all subsets of sector 11's aggregate, not additions to it.
- T-11-06: corporate and strategic investors account for 87.9% of US AI venture deal value, so a large share of "venture funding" is also hyperscaler capex appearing a second time.
Operating rules:
- Never add sector 11 funding totals to any vertical's funding figure. The correct
treatment is
11 = Σ(verticals) + non-venture private capital, with overlap. - Never treat sector 11's growth as independent evidence about any vertical. T-11-02's concentration finding is sector 01's funding story restated at portfolio level.
- Sector 11 has the third-highest out-degree in the map (9) and the second-lowest in-degree (2). That asymmetry is an artefact of it being a layer, not a sector. It is reported separately from the real-economy dependency ranking in §3 for that reason.
- The one genuinely endogenous sector-11 finding — the DPI drought (T-11-01, 0.05x) — is a property of the financing layer itself and does not double-count anything.
2. The major transmission chains
Ten chains are documented end to end. Each link below has a trend ID and a figure; a link without one is not asserted.
Chain A — AI → compute → power → real estate → industrial construction
The canonical chain. Every link is separately evidenced, and the terminal link is subtractive rather than additive, which is what makes it interesting.
| Link | Mechanism | Evidence |
|---|---|---|
| 01 → 03 | Model demand pulls accelerator, memory and packaging demand | T-01-03 inference overtakes training; T-03-01 Micron at 84.9% gross margin, SK hynix at 76% operating margin; T-03-02 CoWoS, not wafers, is the ceiling; T-03-04 TSMC capex to $60–64bn |
| 03 → 01 (return link) | Packaging and HBM allocation caps how many accelerators exist | T-03-02: "accelerator supply forecasts built on foundry wafer capacity are wrong by construction" |
| 01 → 05 | Compute buildout becomes grid load | T-05-01: NERC raised its ten-year summer peak increase to 224 GW, 69% above the prior year; PJM went from 0.3%/yr (2021) to 3.6%/yr |
| 05 → 01 (return link, and the binding one) | Interconnection, not silicon, sets the pace of compute supply | T-01-05: FERC Section 206 show-cause orders to all six US RTOs/ISOs on 2026-06-18; capacity now disclosed in GW (CoreWeave 1.5 GW live, 4.2 GW contracted; Microsoft 38 GW by 2032) |
| 05 → 05 (internal amplification) | Load growth raises prices, which becomes a macro and political variable | T-05-02: PJM wholesale costs +46% to $56.7bn in seven months; FOMC named the AI buildout as an inflation driver; T-05-04 PJM cleared at the $325/MW-day cap and still fell 6,831 MW short |
| 01/05 → 18 | Power availability becomes a real-estate variable | T-18-01: data-centre vacancy 1.4%, Northern Virginia 0.24%; T-18-14 frontier markets displacing established clusters on power and permitting; T-18-11 interconnection queue position emerging as a financeable asset separable from the land |
| 18 → 09 | Data-centre construction bids away crews, switchgear and transformers | T-09-14: data-centre starts $84.1bn YTD, nearly triple, average project cost $352.8m → $1.06bn, cost/sq ft +57% to $818.20; T-18-08 construction unemployment at a record-low 3.1% in a soft market |
| 09 → 09 (terminal, subtractive) | Industrial construction contracts | T-09-02 / T-18-18: private manufacturing construction −21.7% to $167.8bn SAAR; data centres drove all nonresidential growth in July 2026 while non-data-centre spending fell to its lowest since September 2023 |
The chain's defining property is that it is not uniformly additive. The conventional reading is "AI capex is good for industrials." The construction data says AI capex is bidding away the inputs industrial construction needs. Sector 09's own overhyped record (T-09-19) makes the same point on the revenue side: the industrial vendors posting the strongest AI-attributed growth are growing on data-centre demand — Siemens data-centre revenue +35%, US orders +54% — so "industrial AI market" sizing off vendor segment growth double-counts sector 01's capex.
Two secondary branches off Chain A:
- 05 → 21: the marginal megawatt now has a better-paying customer than bitcoin. T-21-18: Cipher Mining renamed itself Cipher Digital; TeraWulf signed a 20-year, 410 MW AI lease. Second-order consequence nobody in sector 21 is discussing: bitcoin's security budget is increasingly funded by companies whose primary business is no longer mining.
- 01 → 07 → 02: AI valuations reach credit. T-07-14 records that 25–26% of direct lending portfolios sit in software and outstanding SaaS loans went from ~$8bn (2015) to over $500bn by end-2025 — ~19% of total direct loans, underwritten against ARR rather than EBITDA, roughly 70% covenant-lite. T-01-02: hyperscaler needs of ~$2.9tn for 2025–2028 at ~60/40 equity/debt, with private credit taking ~$800bn, and ~$970bn of lease commitments of which ~$660bn is off balance sheet.
Chain B — Trade law → retail margins → fashion and industrial earnings → the 2027 cliff
| Link | Mechanism | Evidence |
|---|---|---|
| Court → 09 | The Supreme Court voided IEEPA on 2026-02-20 (6–3); the architecture was rebuilt on Section 232/301 | T-09-01: CBP certified roughly $107bn of refunds by 2026-08-21 |
| 09 → 12 | Refunds land in retail cost of sales | T-12-03: Target booked $994m — 3.7pp of both gross and operating margin and ~$1.65 of FY EPS; Walmart attributed its 96bp gross-margin gain "primarily" to refunds |
| 09 → 23 | The same refunds land in fashion and beauty | T-23-01: NIKE $986m recovery adding ~900bp to fiscal Q4 gross margin and $0.52 of $0.72 EPS; lululemon $134.5m (560bp, $0.86 EPS) in a quarter when comparable sales fell 9%; e.l.f. ~1,050bp of a 1,400bp expansion; PUMA €15.4m; Estée Lauder $38m; adidas excluding a potential $250–300m |
| 09 → 12/23 (second instrument) | De minimis closes on both sides of the Atlantic | T-12-04: US CIT upheld the suspension 2026-08-13; EU €3/item from 2026-07-01 on 5.9bn items, 91% from China. T-12-16: PDD revenue growth fell to +8% with net income −12% |
| → 2027 | Arithmetic | T-23-16: lululemon's underlying gross margin fell ~360bp once refunds are removed. Every company above faces a 2027 comparison it cannot repeat. Mechanism: arithmetic. Compounded in fashion by T-23-14, a forecast 22% increase in the US upland cotton farm price with world ending stocks at a fifteen-year low, landing exactly when refunds stop |
This chain is the clearest example of a court decision propagating into three sectors' reported earnings inside two quarters, and of a cross-sector analysis being strictly necessary: a sector-12 analyst sees a margin story, a sector-23 analyst sees a different margin story, and only the cross-sector view shows they are the same legal event with a shared expiry.
Chain C — Rare-earth licensing → robotics → automotive and defence
| Link | Mechanism | Evidence |
|---|---|---|
| China policy → 09 | Export licensing became a discretionary throttle | T-09-05: US yttrium imports 17 tonnes Apr–Dec 2025 against 333 tonnes in the comparable prior period (−95%); October 2025 extension added a foreign-direct-product rule and a technology/personnel embargo |
| 09 → 13 | NdFeB magnets are ~80% Chinese and sit in every servo motor | Japan received zero covered rare-earth exports in July 2026, while South Korea became the largest destination, above pre-control levels. Japan is home to the world's dominant precision-reducer suppliers and a large share of global robot production |
| 13 → 10, 13 → 08 | Robots and actuators feed vehicle assembly and defence | T-13-17: US automotive robot installations 13,500 units, −1%, while total US installations rose 11% |
| Upstream tell | The real chokepoint is one layer further up than anyone watches | T-13-10: Harmonic Drive moved to the TSE Prime Market in February 2026; Schaeffler will mass-produce strain-wave gearboxes from 2027. Harmonic Drive's consolidated bookings are a better leading indicator of genuine humanoid volume than any humanoid company's announcements |
Why this chain is systematically missed, per sector 13's own dossier: rare-earth licensing is covered intensively as a semiconductor and defence story and almost never as a robot story, because it requires joining a trade-policy dataset to a robotics supply chain and the two literatures do not overlap.
Chain D — AI training rights → publishing and music → AI labs (a reverse-flow chain)
Most chains run from upstream capability to downstream application. This one runs backwards: a downstream sector sets a cost that binds the upstream one.
| Link | Mechanism | Evidence |
|---|---|---|
| 01 → 24 | Training corpora built on unlicensed material | T-24-01: Bartz v. Anthropic final approval 2026-07-20 — $1.5bn, ~500,000 books, ~$3,000 gross per work, 92.77% claims rate, with an order to destroy the torrented files |
| 24 → 01 (the return link, and the binding one) | A settlement becomes the sector's only observed clearing price and makes corpus provenance the dominant legal risk | The settlement was driven by the piracy of acquisition, not by the act of training. Follow-on docket: publishers and authors v. Google over Gemini (2026-07-10); Authors Guild summary-judgment motion against OpenAI and Microsoft (2026-09-04, MDL 25-md-3143) |
| 24 → 24 (distributional) | Who actually gets paid is decided by a formality | T-24-08: of 864 books with compliance data, ~93% showed publisher failure to register copyright, though 54% were under contracts requiring it. Bartz eligibility required registration, so a large share of authors were excluded from a $1.5bn fund by an administrative omission |
| 01 → 24 → 16 | AI-mediated search breaks the referral economy | T-24-05: Pew found users clicked a traditional result on 8% of visits with an AI summary against 15% without, clicked a source inside the summary on 1%, and ended the session on 26% against 16% |
| Market structure | The settlement is bilateral, not collective | T-24-02: Udio-UMG, Udio-WMG, Suno-WMG, Suno-BMG, Suno-Believe — all confidential, no terms disclosed. T-24-20: IFPI is actively lobbying against a statutory right |
Chain E — Middle East conflict → energy → freight and aviation → food → consumer
The single event with the widest documented cross-sector footprint in the database.
| Link | Evidence |
|---|---|
| Hormuz closure → 09 maritime | T-09-07: two chokepoints moving in opposite directions at once; Hapag-Lloyd took a ~$600m Q2 hit with six vessels trapped. T-09-16: rates went from $2,107/40ft in January to $4,476 on 2026-09-10, and Maersk raised full-year EBITDA and EBIT guidance by ~$2bn — invalidating the February 2026 bear case by event, not by modelling error |
| → 09 air cargo | T-09-18: IATA cut 2026 air cargo growth from 2.6% (March) to 0.2% (June) on jet-fuel and Gulf-hub disruption, after February CTKs of +11.2% |
| → 25 aviation | T-25-02: jet fuel at 31.4% of operating cost against 25.4%; industry net profit halved to $23bn; Delta paid $3.93/gal (+75%), United $4.19 (+79.4%) |
| → 05 power | T-05-02: Ofgem raised the GB October 2026 cap 4% to £1,723, naming Middle East conflict-driven wholesale gas |
| → 19 food | T-19-03: European gas at $21.11/mmbtu against US Henry Hub $2.77 — a 7.6x spread — making European ammonia structurally uncompetitive; the Commission mobilised €540m from the agricultural reserve under a "Middle East Crisis Temporary State Aid Framework" |
| → 25 demand | T-25-06: Middle East travel market contraction as direct conflict transmission |
One event, five sectors, all within two quarters. No sector-level research design would have caught it; the macro brief is what made it visible.
Chain F — Chinese industrial policy → cells → vehicles → European OEMs → robotics
| Link | Evidence |
|---|---|
| China policy → 10 cells | T-10-08: LFP at $81/kWh against NMC $128/kWh; China pack prices $84/kWh against North America $121 and Europe $131 — a 44–56% cost penalty outside China. Seven Chinese firms hold 72.8% of the top-ten cell market, +3.1pp |
| 10 cells → 10 OEM | T-10-01: China exported 7.15m vehicles Jan–Aug 2026 (+66.7%), of which 3.44m NEVs (+120%), while domestic NEV retail fell 10.1% y/y in August, the eighth consecutive monthly decline |
| 10 → Europe | T-10-02: Chinese brands at ~10% of European new-car sales and ~13% in the UK; PHEV sales rose ~14x in a year to arbitrage the 35% BEV duty. T-10-18: Stellantis took €25.4bn of charges; VW's four ID plants under review to 2031 |
| 10 → 13 | T-13-01: China's domestic robot makers moved from 30% to 57% of their home market between 2020 and 2024; T-13-02: global installations plateaued near 540,000 units with China absorbing all the growth |
Chain G — App-store law → gaming and consumer apps → creator economy
| Link | Evidence |
|---|---|
| Courts/statutes → 15/17 | T-15-01: the flat 30% commission dismantled across seven jurisdictions in nine months. Apple EU from 2026-10-01: 26%/20%/15% plus a 5% Core Technology Commission; Japan MSCA 21% + 5%; China 25%/12% from 2026-03-15 — the first unilateral, non-court-ordered cut in a major market |
| 15 → 17 → 15 | T-17-03: age verification migrating to the app-store and OS layer (Apple's Declared Age Range API, Significant Change API, Social Media Time Allowance category, required September 2026). Whoever controls the age signal controls conversion funnels and default screen-time limits for every consumer app |
| Cost incidence | The pro-competitive remedy at the top of the stack is a fixed compliance cost at the bottom: per-jurisdiction fee agreements, age assurance, AI disclosure and child-safety architecture all land on developers |
Chain H — Finance → everything, through two specific mechanisms
Sector 07 exports risk rather than product. Two named transmission mechanisms, both under-covered:
- 07 → 02 (software credit). T-07-14: ~25–26% of direct-lending portfolios in software; SaaS loans ~$8bn (2015) → $500bn+ (end-2025); ~70% covenant-lite, so the signal arrives only at payment default. This is the specific, nameable mechanism by which the FOMC's AI-valuation concern becomes credit losses.
- 07 → 11 → retail investors. T-11-07 evergreen AUM approaching $500bn with wealth investors at one-fifth; T-11-16 the structure gated in Q1 2026 (Blue Owl OBDC II closed redemptions; Cliffwater received requests for 14% against a 7% cap).
Return link 21 → 07: T-07-20 records the stablecoin deposit-disintermediation claim as an advocacy estimate, against FDIC-observed domestic deposit growth of +0.8% in Q2 2026, an eighth consecutive increase.
Chain I — Climate policy retreat → software and assurance TAM
A chain that runs from regulation directly into two vendors' addressable markets and is almost entirely uncovered.
| Link | Evidence |
|---|---|
| 20 → 02 | T-20-02: Directive (EU) 2026/470 narrowed CSRD to undertakings exceeding €450m turnover and 1,000 employees, and created "protected undertakings" that may refuse value-chain information requests, with any contrary contractual provision not binding. The addressable market for EU sustainability reporting software fell by roughly an order of magnitude in company count, and the Scope 3 collection model most carbon-accounting products are built on became legally unenforceable |
| 20 → audit | T-20-16: Article 26a(3) and its October 2028 reasonable-assurance deadline were deleted. Limited assurance is permanent; the Big Four staffed for a revenue event that is not coming |
| 20 → 05 | T-05-16: OBBBA ended clean-energy credits on 2026-07-04; solar and wind interconnection requests each −19% in the 2026 queue |
| 20 → all (data) | T-20-03: if EPA removes GHGRP obligations for 46 source categories, "the calibration set under most US climate analytics degrades permanently and the products silently become models rather than measurements" |
Chain J — The measurement layer (a dependency, not a causal chain)
Not a flow of value but a shared foundation whose degradation propagates to every sector that reads it. Documented instances:
- US Census reports data centres inside "Office." Census office construction is +21.3% y/y in the middle of an office-distress narrative. Any sector touching nonresidential construction — 09, 18, 05 — must decompose it or be wrong. This single classification choice corrupts three sectors' headline series.
- Core CPI 2.4% (Aug 2026) against core PCE ~3.3% — an unusually wide gap in the unusual direction, unreconciled, and relied on by 12, 19, 07 and 05.
- Effective-tariff-rate conflation: Penn Wharton 6.7% "effective" vs Tax Foundation 7.2% effective / 11.8% "applied" — definitional, not factual, and quoted interchangeably in 09 and 12.
- CBRE vs JLL on data-centre capacity: ~9x (T-18-20), read by 18, 05 and 01.
- Crunchbase vs KPMG on H1 2026 global VC: $510bn vs $560.4bn, and Q2 deal count "5,000+" vs 8,440 — read by 11 and every vertical that cites venture funding.
- 201 recorded contradictions across the database, most densely in 08 (18), 10 (11), 16 (11), 01 (10), 09 (10), 11 (10) and 21 (10).
Chain K — Healthcare pricing → manufacturing → trade
T-06-02 MFN pricing → T-06-01 incretin price/volume reset → T-06-06 / T-06-20 tariff-forced onshoring, where a manufacturing pledge is directly convertible into Section 232 relief (20%, or 0% combined with an MFN agreement) → 09, where the pledges appear as an industrial-construction expectation that Census cannot find. The same instrument therefore produces a drug-pricing outcome, a tariff outcome and a phantom construction pipeline.
3. Dependency ranking
Constructed from the 79 documented directed links above. Out-degree counts the sectors a sector demonstrably drives; in-degree counts those that drive it. Net degree is the upstream-ness score.
3.1 Real-economy ranking (sector 11 excluded — see §1)
| Rank | Sector | Out | In | Net | Interpretation |
|---|---|---|---|---|---|
| 1 | 01 AI & compute | 12 | 5 | +7 | The most upstream sector in the database. Changes originate here and propagate to at least twelve others. Also the sector with the lowest data availability score (3/5) in the industry ranking — maximum influence, minimum observability |
| 2 | 05 Energy & power | 8 | 2 | +6 | Second-most upstream and the binding one. Where 01 creates demand, 05 decides whether it can be served. Highest data availability (5/5), so the most auditable upstream sector |
| 3 | 09 Industrial / supply chain | 9 | 6 | +3 | The trade-policy and logistics transmission hub. Touches more sectors than any other (15 distinct partners) but is itself heavily driven |
| 4 | 07 Finance | 5 | 2 | +3 | Exports risk, not product. Its influence is under-mapped because it arrives as a credit event, not a product event |
| 5= | 03 Semiconductors | 4 | 3 | +1 | Ranked #1 on the industry scorecard (97.4) and cross_industry_influence 5/5, but bidirectionally coupled to 01, which is why its net position is lower than its importance |
| 5= | 16 Advertising | 4 | 3 | +1 | Sets the revenue model for 12, 14, 17 and 24 |
| 5= | 20 Climate | 3 | 2 | +1 | Small sector, disproportionate regulatory reach into 02, 05 and 09 |
| 8 | 06 Healthcare | 3 | 3 | 0 | Drives 09, 19 and 23 through pricing and pledges |
| 8 | 14 Media, 08 Aerospace, 15 Gaming, 22 Education | 1–2 | 1 | ~0 | Balanced |
| 12 | 04 Cybersecurity | 2 | 3 | −1 | A cost layer across all sectors; genuinely downstream |
| 13 | 17 Social, 25 Travel | 1 | 2 | −1 | |
| 15 | 23 Fashion, 24 Publishing | 1 | 3 | −2 | 24 is the exception that proves the rule: one outbound edge, and it is decisive (Chain D) |
| 17 | 12 Retail, 13 Robotics, 10 Automotive, 19 Food, 21 Crypto | 1–3 | 4–6 | −3 | Terminal demand and terminal assembly sectors |
| 22 | 18 Real estate | 1 | 5 | −4 | Almost purely a receiver of AI, energy and capital decisions |
| 23 | 02 Enterprise software | 1 | 6 | −5 | The most downstream sector in the database. Driven by 01, 04, 07, 11, 20 and 22 simultaneously — which is why its 2026 equity de-rating (T-02-03, ~$1tn of value) was set by a narrative about sector 01 rather than by its own metrics |
3.2 The capital layer, reported separately
| Sector | Out | In | Net |
|---|---|---|---|
| 11 VC & private markets | 9 | 2 | +7 |
Nominally tied with sector 01 for most upstream. This is an artefact and must not be read as influence. Sector 11 does not cause change in nine sectors; it finances change that is caused elsewhere, and its metrics restate theirs. See §1.
3.3 How this compares to the analyst-assigned scores
The industry ranking's cross_industry_influence criterion (assigned per sector during
scoring, before this map existed) gives 5/5 to sectors 01, 02, 03, 05, 07, 09 and 11.
The derived map agrees on 01, 03, 05, 07, 09 and 11 and disagrees sharply on 02, which
the derived map puts last. The disagreement is informative: analysts scored enterprise
software as influential because its products are everywhere, while the link evidence shows
that change in it originates elsewhere. Ubiquity is not upstream-ness. Phase 2 should
redefine the criterion accordingly.
4. Adjacency table
Directed. Read a row as "this sector drives →". Sectors ordered by net degree.
C marks a link that is part of a named chain in §2.
| ↓ drives → | 01 | 02 | 03 | 04 | 05 | 06 | 07 | 08 | 09 | 10 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 01 AI | — | C | C | C | C | C | C | C | C | C | C | C | ||||||||||||
| 05 Energy | C | — | C | C | C | C | C | C | C | |||||||||||||||
| 09 Industrial | C | C | — | C | C | C | C | C | C | C | ||||||||||||||
| 07 Finance | C | — | C | C | C | |||||||||||||||||||
| 03 Semis | C | — | C | C | C | |||||||||||||||||||
| 16 Advertising | C | C | — | C | C | |||||||||||||||||||
| 20 Climate | C | C | C | — | ||||||||||||||||||||
| 06 Healthcare | — | C | C | C | ||||||||||||||||||||
| 14 Media | — | C | C | |||||||||||||||||||||
| 08 Aerospace | C | — | C | |||||||||||||||||||||
| 13 Robotics | C | C | — | |||||||||||||||||||||
| 04 Cyber | C | — | C | |||||||||||||||||||||
| 12 Retail | — | C | C | C | ||||||||||||||||||||
| 15 Gaming | — | C | ||||||||||||||||||||||
| 17 Social | C | — | ||||||||||||||||||||||
| 10 Automotive | — | C | ||||||||||||||||||||||
| 18 Real estate | C | — | ||||||||||||||||||||||
| 19 Food | C | — | ||||||||||||||||||||||
| 21 Crypto | C | — | ||||||||||||||||||||||
| 22 Education | C | — | ||||||||||||||||||||||
| 23 Fashion | C | — | ||||||||||||||||||||||
| 24 Publishing | C | — | ||||||||||||||||||||||
| 25 Travel | C | — | ||||||||||||||||||||||
| 11 VC (layer) | ✚ | ✚ | ✚ | ✚ | ✚ | ✚ | ✚ | ✚ | ✚ |
✚ = financing relationship, not an independent causal link. Never summed with the target sector (§1).
5. Five practical consequences
Upstream sectors are the least observable. Sector 01 has the highest out-degree (12) and a
data_availabilityscore of 3/5 — the lowest among the top four ranked sectors. No frontier AI lab discloses a gross margin. The sector with the most propagating influence is the one where independent verification is weakest, and Epoch AI's inference price-trends dataset — the most-cited independent source on the cost curve — was last updated 2025-03-12, eighteen months before this research date.The binding constraint is always one sector downstream of where attention sits. Attention is on models (01); the constraint is interconnection (05). Attention is on accelerators (03); the constraint is packaging (03, but a different layer). Attention is on humanoids (13); the constraint is precision reducers (13, upstream, Japanese duopoly). Attention is on Golden Dome interceptors (08); the constraint is solid rocket motors (08, upstream, two suppliers, 30-month lead time). The generalisable heuristic: find the input with two suppliers and a lead time.
Regulatory and legal events cross sector boundaries faster than commercial ones. The IEEPA ruling reached three sectors' reported margins in two quarters (Chain B). The Hormuz closure reached five sectors in the same window (Chain E). No commercial development in this database propagated that fast.
The most under-mapped edges are the reverse-flow ones. 24 → 01 (training-rights pricing binding AI labs), 20 → 02 (climate deregulation destroying software TAM), 05 → 01 (interconnection capping compute), 10 → 13 (the end of EV retooling ending robot demand growth). Each runs from a lower-ranked sector into a higher-ranked one, and each is invisible to a research design organised by sector prominence.
Two sectors are structurally mislabelled in the current taxonomy. Sector 11 is a capital layer (§1). Sector 04 is a cost layer — it appears in every sector's risk register and drives almost nothing on its own. Both should be typed differently in Phase 2 so that cross-sector aggregation is arithmetically safe by construction rather than by analyst discipline.
- Source artifact
- 04-analysis/43-cross-industry-map.md
- Corpus date
- 15 September 2026
- Prepared for this site
- 16 September 2026
- Site publication
- 18 September 2026
- Verification
- Inherited; not fully rechecked