The 44,000-Company Supply Chain Map That Cost $0 to Build
Here is the most expensive piece of information your company buys that should now be free: your supply chain map.
Most large corporations spend between $500,000 and $5 million annually on proprietary databases — Bloomberg SPLC, IHS Markit, S&P Global, Capital IQ — to understand who supplies whom, which firms are central in their value chain, and where concentration risk lives. These datasets are expensive. They are incomplete. And they are slowly updated — often quarterly at best, with weeks of lag between a real-world event and its reflection in the database.
A research team from the Complexity Science Hub Vienna, the Medical University of Vienna, and the Austrian Institute of Economic Research just proved that this entire industry is now obsolete.
They used large language models to crawl archived corporate websites, press releases, and investor relations pages — publicly available web data, no paywalls, no subscriptions — and reconstructed the global production network of the semiconductor industry. The result: a map of 44,000 firms connected by 2.1 million unique relationships, spanning 1,145 cities and 64 countries, covering 25 years of structural evolution.
The cost of the data: zero.
The cost of the proprietary equivalent: millions.
For anyone responsible for corporate strategy, investor relations, supply chain resilience, or M&A, this paper is not a technology curiosity. It is a demonstration that the information monopoly on supply chain intelligence has been broken.
Cost of the data — 44,000 firms, 2.1M relationships, 25 years
vs. $500K–$5M annually for proprietary database licenses
Executive Summary
The study: Researchers used LLMs (GPT-4o-mini) to extract firm-to-firm relationships from open web data — no proprietary databases, no paywalls, no surveys. Validated on the semiconductor industry, seeded with 124 firms, expanded to 44,000+ firms through iterative crawling.
Surge from 2022
Among 44,000 firms, one centrality line dominates. NVIDIA’s betweenness centrality exploded in 2022 with ChatGPT’s launch. No other firm shows comparable shift.
Fabless + Foundry Rise
Network captures the rise of fabless chip designers (AMD, Qualcomm, Apple) and the growing centrality of contract manufacturing (TSMC). All from public data.
Validated vs S&P Global
67% agreement with proprietary S&P data. 89% human annotation precision. Year-by-year temporal granularity that no proprietary database provides.
Multi-level capability
Centrality measures, community detection, geographic concentration, temporal evolution. A dynamic model of your competitive environment from free data.
The era of closed, expensive, proprietary supply chain databases is ending. Any company can now map its supply chain dependencies, identify bottleneck firms, and track structural shifts — all from public data. For corporate strategy and investor relations, this is the equivalent of GPS replacing paper maps.
Paper at a Glance
| Metric | Value |
|---|---|
| Title | Reconstructing Temporal Multi-Relational Firm Networks at Scale Using Large Language Models: The Case of the Semiconductor Industry |
| Authors | Köse, Diem, Dervic, Friesenbichler, Heiler, Koller, Klimek |
| Institutions | Complexity Science Hub Vienna, Medical University of Vienna, Austrian Institute of Economic Research |
| Published | May 15, 2026 |
| Categories | physics.soc-ph, cs.SI, q-fin.GN |
| Relevance Score | 96/100 — Validates open-data supply chain reconstruction across industries |
| Paper URL | arxiv.org/abs/2605.15842 |
What the Paper Found
Finding 1: NVIDIA’s Centrality Surge Is Visible in Open Data
Among 44,000 firms tracked between 2015 and 2025, one line on the chart tells the story of the last three years in the global economy. NVIDIA’s betweenness centrality — a network measure that captures how much of a bottleneck a firm is in the flow of goods, services, and relationships between other firms — was relatively stable until 2022. Then it exploded.
The surge coincides with ChatGPT’s commercial launch in November 2022 and the AI compute boom. NVIDIA’s centrality trajectory is so dramatic that it visually dominates the chart of all 44,000 firms. No other firm — not TSMC, not Samsung, not Intel — shows a comparable shift.
So what: The market has been pricing in NVIDIA’s dominance. This paper provides independent, open-data confirmation that NVIDIA has become the structural bottleneck of the AI economy. Every AI company, every cloud provider, every automotive firm building AI features routes through NVIDIA’s supply chain.
Finding 2: Semiconductor Industry Structural Shifts Captured in Open Data
The network reconstruction reveals two major structural shifts consistent with known industry narratives:
- The rise of fabless firms. Companies that design chips but do not manufacture them — AMD, Qualcomm, Apple — have grown substantially in network centrality.
- The growing role of contract manufacturing. TSMC’s centrality has grown consistently, highlighting the concentration risk of the foundry model.
So what: The structural dynamics of your industry can now be mapped from public data. Whether you are in semiconductors, automotive, pharmaceuticals, or any industry with complex supply chains, the same methodology applies.
Finding 3: Open Data Replaces Proprietary Databases with Comparable Accuracy
The 67% agreement rate with S&P Global data is higher than typical agreements between different proprietary databases. The 89% human annotation precision exceeds the accuracy of many manual supply chain mapping exercises. And the temporal granularity — year-by-year relationship tracking — is something no proprietary database provides.
So what: Your annual $500K+ expenditure on supply chain data vendors should be reviewed. The capability gap between what you pay for and what you can get for free is narrowing rapidly.
Finding 4: The Network Supports Strategic Analysis at Multiple Levels
The reconstructed network enables analyses proprietary databases struggle with:
- Centrality measures — identify bottleneck firms that would cascade through the value chain if disrupted
- Community detection — identify clusters of closely connected firms functioning as de facto ecosystems
- Geographic concentration — 1,145 cities across 64 countries showing exactly where supply chain risk lives
- Temporal evolution — year-by-year tracking over 25 years
So what: This is a dynamic model of your competitive environment that updates as new web data becomes available.
Why This Matters for Executives
Chief Strategy Officers & Heads of Corporate Development — Your competitive intelligence just received a massive zero-cost upgrade. Action: Commission a proof-of-concept for your industry. Identify the 20 most strategically important firms and map their relationship networks from public data. Compare to your proprietary databases — the gap reveals what your competitors can see that you cannot.
Heads of Investor Relations & CIOs — Firm-level centrality is a leading indicator of systemic importance. NVIDIA’s centrality surge began in 2022 — before the full market pricing of the AI compute boom. Action: Add network centrality to your investment screening and portfolio monitoring process.
Chief Supply Chain Officers — Supply chain visibility has always been the holy grail. This method extends visibility to all tiers. Action: Run your company’s supply chain through this methodology. Identify Tier 2 and Tier 3 dependencies you were not tracking. Map concentration risk.
Heads of M&A — Network proximity predicts acquisition success better than industry classification. Structurally adjacent firms integrate more successfully. Action: Add network distance to your M&A target scoring framework.
How This Fits the Series
| Date | Category | Paper Topic |
|---|---|---|
| May 16 | Customer Service & CX Automation | Agentic AI Field Experiment on Taobao (Alibaba) |
| May 17 | PR & Communications | Google AI Overviews Audit (Princeton/Syracuse) |
| May 18 | Corporate Strategy & Supply Chain Intelligence | LLM Firm Network Reconstruction (This Paper) |
New business function: Corporate Strategy & Investor Relations (AI-Driven Supply Chain Intelligence) — 59th business function covered. First paper on open-data supply chain intelligence in the series.
Conclusion
The information monopoly on supply chain intelligence has been broken. Large language models can now reconstruct the hidden wiring of the global economy from publicly available web data — with accuracy comparable to proprietary databases costing millions. The single most dramatic signal in the reconstructed network — NVIDIA’s centrality surge — validates both the methodology and the narrative.
For every corporate strategy, investor relations, supply chain, and M&A executive, this paper is a wake-up call. The question is no longer whether open-data supply chain intelligence works. The question is whether your competitors are using it while you are still paying for what is now free.
“Our approach overcomes key limitations of proprietary firm-level datasets, offering a scalable, open, and continuously-updatable methodology for mapping production networks in any industry.”
— arXiv:2605.15842, Complexity Science Hub Vienna
Frequently Asked Questions
What does this mean for a Chief AI Officer?
You now have proof that LLMs can solve high-value, previously proprietary business problems at near-zero cost—validating investment in AI infrastructure and internal capability building. This research demonstrates that competitive advantage shifts from purchasing expensive third-party databases to deploying AI systems that extract intelligence from public data, making your AI strategy directly material to corporate strategy.
How should supply chain teams use a 44,000-firm map to identify concentration risk differently than proprietary vendors?
Real-time visibility into 2.1 million relationships lets you spot emerging bottlenecks like NVIDIA’s centrality in AI supply chains months before traditional quarterly updates surface them. You can now run continuous scenario analysis on supplier dependencies rather than waiting for annual reports, fundamentally shifting risk management from reactive to predictive.
Should our company conduct an AI Assessment for companies to evaluate internal LLM supply chain capabilities?
Absolutely—understanding whether your organization can replicate this research internally versus continuing to pay millions for proprietary data is now a core competency gap. Silicon Valley Certification Hub offers AI Assessment for companies specifically designed to map your technical readiness to build intelligence systems like this, helping you avoid legacy vendor lock-in.
What should our executive team do in the next 90 days with this research?
First, audit which supply chain, strategy, and risk teams still rely on expensive proprietary databases—then initiate a proof-of-concept with your data science team to test LLM-based reconstruction on your specific industry or value chain. Second, evaluate whether building this capability internally or partnering with research institutions makes sense given your competitive positioning and the demonstrated zero marginal cost of data.
0 Comments