Scrutica
Where NVIDIA sits in the compute supply chain — geographic revenue split, segment mix, capex / FCF / buyback history, and industry placement — sourced from licensed-database extracts and SEC EDGAR, joined to the supply-chain graph through the industry classification.
Per data-quality principle §7 (CONTRADICTION SURFACING): when primary sources disagree under different methodologies, both are shown with as-of dates rather than picking a silent winner.
supplier_graph_density_anomalyNvidia's licensed supply-chain 2026-05-14 surfaces 651 disclosed counterparty edges anchored to NVDA-US, including direct customer relationships with Gigabyte Technology (TW; 3M price correlation 0.87) and IEIT Systems A (CN; rank 100, 3M price correlation 0.93) and partner relationships with Indosat (ID) and Vobile (HK). However, the entities Culper Research 2026-05-13 names as channelling the largest portion of Southeast Asian Nvidia volume — Megaspeed International Pte. Ltd., Speedmatrix Sdn. Bhd., Apex Enterprise Solutions, Aolani, Novagate Cloud, Siam AI, OBON Corp, INF Tech, and the post-rebrand Aivres Systems Inc. — do NOT appear in the disclosed customer / partner / supplier graph at any rank. The absence is itself the finding: a 651-edge graph with 254 disclosed customers should reasonably surface the largest Southeast Asian buyer if that buyer were transacting directly with Nvidia. Culper's thesis is that the volume is routed through OEM intermediaries (Aivres for ~$4B; Giga Computing for ~$500M) — which IS consistent with the licensed corporate-ownership/supply-chain database graph showing direct Nvidia-to-OEM relationships but no Nvidia-to-end-buyer relationships.
Per DQP §7, the licensed corporate-ownership/supply-chain database's supply-chain graph is structurally correct: it surfaces the direct OEM relationships Nvidia discloses (Aivres / Inspur Information, Gigabyte / Giga Computing, Supermicro). What it does NOT surface — and structurally CANNOT surface — is the chain of intermediaries downstream of the OEM. The flag exists so researchers reading the Nvidia supplier-graph rendering understand the surface they are looking at is the upstream-half of the channel, not the full diversion path Culper alleges.
geographic_revenue_methodology_change_eventPer Culper Research 2026-05-13 citing Nvidia's FY Q3 2026 (October 2025) disclosure: Nvidia changed its geographic-revenue reporting methodology from "customer billing location" to "customer headquarters location" — eliminating the entirety of what the company had previously reported as Singapore-based revenue. The 10-Q for the period explains the change: "we believe changing to revenue based upon the location of our customers' headquarters provides a better representation of the geographic profile of our revenue." The timing is structurally significant: Singapore had grown from 8.5% of revenues in FY 2023 to 20.1% of revenues for the LTM ended July 2025 — over 99% of which Nvidia maintains is tied to U.S. end customers using Singapore-based "centralized billing practices." licensed geographic-revenue 2026-05-14 records LTM Jan '26 revenue of USD 215.9B, 69.3% United States (+47.6% Y/Y), 19.6% Taiwan (+24.4%), 8.9% Mainland China (-30.5% Y/Y), with Singapore now absent from the top 8 — consistent with the methodology shift.
Per DQP §7, both attribution methodologies (billing-location and HQ-location) are defensible. The flag exists so a researcher comparing pre-Q3-FY2026 to post-Q3-FY2026 geographic-revenue disclosures understands the discontinuity is methodological, not economic — and so that the apparent compression of Singapore exposure is read in context of the reported attribution change rather than as a real volume decline.
| Year | Capex | Op. CF | FCF | Buybacks |
|---|---|---|---|---|
| FY2026 | ||||
| FY2025 | ||||
| FY2024 | ||||
| FY2023 | ||||
| FY2022 | — |
Dollar flows are annual figures from the fiscal year labelled on the row — SEC segment and >10%-customer disclosures, re-derived company disclosures, or a licensed supply-chain database; per-row source and authority tier on hover. A row without a year means the source does not state the period. The 3-month Pearson r comes from a licensed supply-chain database; correlation is a coupling proxy — shared market sensitivity, not a causal channel. Sole-source edges are flagged inline.
The substrates above join to the platform's analytical surfaces. Each link carries NVIDIA's own numbers into the tool that acts on them.
436 upstream · 397 downstream — the interactive graph opens with NVIDIA selected. The cascade simulator propagates seed disruptions across this same edge set, weighted by the curator's substitutability assessment.
The facility table filtered to exactly these records — class, status, and power per row, exportable with full provenance columns.
Facilities, sovereign-AI programs, export-control posture, and the announced-versus-deployed reality gap for the country.
Side-by-side facilities, GPU inventory, power, revenue, capex, and market cap.