A Knowledge Graph Can Show a Connection. Supply-Chain Data Shows the Economic Dependency.
September 14, 2026
Altsets
Research by Altsets Research
A generic knowledge graph can connect two companies. Investment-grade supply-chain data adds supplier-customer direction, economic importance, explicit missingness, point-in-time history, and security mapping so the connection can support an actual research decision.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- The supplied 561M USD HPE-Microsoft relationship illustrates how direction and economic magnitude turn a generic company connection into a ranked investment-research dependency.
- The supplied Tesla-SpaceX example shows why relationship type must remain explicit because a commercial customer relationship and an equity investment support different conclusions even when the same two companies appear in both.
A generic knowledge graph can tell you that two companies are related. Supply-chain data becomes more useful for investing when it tells you which company supplies the other, how economically important the relationship is, whether the metric is known or missing, and what the relationship looked like at a historical point in time. The difference is not graph versus graph. It is generic entity connectivity versus a graph built around a specific economic question.
"Related to" is not enough for an investment conclusion
Google describes a knowledge graph as a collection of real-world entities represented as nodes. That structure is powerful for entity search and semantic relationships. For investing, however, knowing that two organizations are connected is only the beginning. The direction and economic meaning of the connection determine what the investor can actually do with it.
The supplied Altsets data contains a 561M USD relationship from Hewlett Packard Enterprise to Microsoft. That direction matters. It tells the researcher that Microsoft is a customer in the mapped commercial relationship rather than merely another organization appearing near HPE in a graph. The same relationship can then be used to ask whether Microsoft earnings, enterprise spending, or hybrid-cloud developments deserve attention when researching HPE.
Direction changes the question
A generic company graph can contain partnerships, investments, shared executives, litigation, common technologies, customers, suppliers, competitors, and ownership links. Those relationships are not economically interchangeable. A supplier link creates a different hypothesis from a customer link, and both are different from an equity investment or strategic alliance.
The supplied Tesla-SpaceX data illustrates why this distinction matters. Tesla can have a commercial relationship with SpaceX while separately holding an equity investment. Combining those facts into a single generic "Tesla is connected to SpaceX" edge would lose the distinction between operating demand and investment value. A dependency dataset preserves the relationship type so the investor does not silently turn one economic channel into another.
Economic metrics make the graph rankable
A second difference is prioritization. If a stock has twenty connected organizations, a generic graph may establish that all twenty relationships exist. It does not necessarily tell the investor which outside company matters most to the target's revenue, cost base, or commercial activity.
Altsets can carry relationship size, supplier revenue percentage, and customer cost percentage where those metrics are available. The supplied HPE relationships show Microsoft at 561M USD, Swisscom at 203M USD, Home Depot at 109M USD, and Volkswagen at 64.5M USD. Within that displayed four-edge set, Microsoft is clearly the largest relationship by value. The network becomes a research-priority system rather than a flat collection of connections.
The missingness rules matter just as much. A structural edge without an economic metric remains a known relationship without an invented weight. That is a different philosophy from scoring every graph edge simply because an algorithm needs a dense matrix.
Historical state is another layer a generic graph may not preserve
Investment research often asks what was known or economically true at a prior date. Company names change, securities delist, mergers alter corporate boundaries, and relationships appear or disappear. A current entity graph can be completely accurate today and still be unusable for a historical backtest if it cannot reproduce the old relationship state.
Altsets' point-in-time snapshots make the graph a historical research object rather than only a current semantic map. The investor can ask when a relationship existed, whether its measured importance changed, and what security represented the economic entity at the time. That becomes essential once the graph is used for historical portfolio or trading research.
The conclusion is that investment usefulness comes from relationship semantics
A generic knowledge graph is excellent at representing entities and connections. Supply-chain data becomes differentiated when the edge itself carries economic meaning: supplier versus customer direction, comparable relationship metrics, explicit missingness, historical state, and security mapping. The moat is not drawing lines between companies. It is making those lines usable for investment decisions without losing what the relationship actually means.
The public-sources proprietary-data guide explains why normalization and historical reconstruction can remain proprietary even when many source documents are public. The Tesla and SpaceX guide shows why relationship type matters when two companies are connected in more than one way.
For relationship definitions and evidence limits, read the Altsets methodology.