Karp's critique reframes the enterprise AI debate around who controls the operating intelligence of the enterprise — not which model it picks. This collection adds a missing perspective: a Linked-Data-based critique of Systems of Record and the System of Intelligence.
Karp argues frontier model vendors intend to extract enterprise knowledge and erode proprietary advantage — theCUBE calls this Data Communism. Its counter, Data Capitalism, keeps proprietary advantage exclusive to the enterprise and its ecosystem via two stack layers.
The real debate is not open versus closed models — it's who controls the enterprise's operating intelligence. theCUBE lays out two competing scenarios (Frontier Model Leadership vs. Dispersed Intelligence) and a five-layer control-point map of the likely fragmented market. This knowledge graph adds a distinct, independently authored perspective: the article invokes "systems of record," "ontology," and "digital twin" without ever specifying a representational substrate — a gap addressed here through Linked Data and Semantic Web principles.
Who ends up owning the operating intelligence of the enterprise?
Frontier model vendors dominate because their utility, cost curves, research velocity, volume and compute access outpace everything else; enterprises route more work to them as inference costs fall (Wright's Law-style cost modeling).
Frontier players lack the mindset, DNA, process knowledge and trust to own the System of Intelligence; Palantir, Databricks, Microsoft, Google, Celonis and startups instead cement the more critical position.
theCUBE expects the market to fragment across five control points, each with a different set of likely competitors.
Microsoft, OpenAI, Anthropic, Google, Databricks.
Palantir, Databricks, Snowflake, Microsoft, Salesforce, SAP, ServiceNow, Celonis, RelationalAI.
Databricks Unity AI Gateway, Palantir Evolve, Microsoft.
Palantir.
The AI agent authored this distillation on behalf of Kingsley Uyi Idehen, informed by his long-standing public writing on Linked Data and the Semantic Web, applied as an independent critique of the article's treatment of Systems of Record and the System of Intelligence.
HTTP is the open standard that brings Internet connectivity to documents and the entity relationships they contain, of which the World Wide Web is a successful and readily experienced use-case example. This is not a claim about data — it is a claim about information and knowledge: a machine-computable approach to representing the variety of context lenses applied to data.
Enterprises must leverage the power of HTTP to loosely couple AI agents and data spaces (databases, knowledge bases, file systems, and APIs). This is a fundamental best practice for retaining control over the methods and sources that underpin operational success. Those methods and sources should never be outsourced to third-party technology vendors, whether overtly or covertly.
The seven critiques below are applications of this thesis to the article's specific claims.
A real System of Record requires every business entity to carry a globally unique, dereferenceable HTTP identifier (IRI), not merely a primary key scoped to one vendor's database schema.
Palantir Ontology and Databricks Unity Catalog/Genie Ontology are each closed, proprietary graphs — reproducing, one layer below the model, the exact lock-in Karp attributes to frontier labs.
Tim Berners-Lee's four rules describe an enterprise-owned, standards-based Knowledge Graph that any model, frontier or open, can query without a single vendor's proprietary ontology API.
RDF offers an application- and vendor-neutral model, unlike relational tables bound to one application's schema.
The enterprise owning its own Knowledge Graph, rather than choosing between a frontier lab and a vendor intermediary.
SPARQL, an HTTP-protocol extension for declarative querying of knowledge graphs, is inherently model- and data representation format-agnostic, unlike a proprietary model-routing API such as Unity AI Gateway or Evolve.
WebID and W3C PROV-O provenance, not proprietary log rows, should anchor captured agent traces and tacit knowledge in AI agent readable knowledge graphs.
Karp argues frontier vendors intend to extract enterprise knowledge and erode proprietary advantage — "Data Communism." theCUBE reframes the issue as control of the System of Intelligence, not model choice.
Data Communism gives every firm access to the same intelligence. Data Capitalism keeps proprietary advantage exclusive to the organization and its ecosystem.
The SoI captures business rules, policies, processes, state and tacit knowledge as governed assets, shaping inputs to and outputs from any LLM.
Wright's Law-style modeling: as cumulative usage, tokens, feedback, training, and compute rise, costs fall predictably.
So no single frontier vendor can later intermediate, compete with, or extract margin from its enterprise customers.
Palantir is closer to an executable ontology/operational decision layer; Databricks builds up from governed data and semantic context (Genie, Unity Catalog).
Jensen Huang: "proprietary and open," not "versus." The model router becomes the practical control point.
Model utility, engagement, System of Intelligence, agent governance/routing, and managed outcomes.
Authoritative metrics, business definitions, policies/permissions, process logic, decision rights, workflow state, tacit knowledge, and agent traces.
It never specifies a representational substrate — Kingsley Idehen's critique calls for globally unique dereferenceable IRIs per entity.
Palantir Ontology and Databricks Unity Catalog are each closed, proprietary graphs — the same walled-garden dynamic, one layer down the stack.
An enterprise-owned Knowledge Graph built on Linked Data principles, queryable via SPARQL by any model.
HTTP is the open standard that brings Internet connectivity to documents and the entity relationships they contain — the Web is one use case of this, not the only one. This is a claim about information and knowledge, not data: enterprises must use HTTP to loosely couple AI agents and data spaces (databases, knowledge bases, file systems, APIs), retaining control over the methods and sources that underpin operational success rather than outsourcing them, overtly or covertly, to third-party technology vendors.
Twelve terms wrapped in a schema:DefinedTermSet in the companion RDF.
The enterprise-context layer around a frontier model, shaping intent, constraints and governed action.
The client/user surface layer through which enterprises and agents interact with intelligence.
The authoritative data source for enterprise information; in a Linked-Data architecture, its entities carry globally unique dereferenceable identifiers.
A formal specification of domain concepts and relationships — vendor-proprietary (Palantir Ontology) or open, standards-based (RDF/OWL).
A live virtual representation of an enterprise process or asset, part of the operational context layer.
Keeping proprietary data, process and business advantage exclusive to an organization and its ecosystem, as opposed to Data Communism.
Tim Berners-Lee's principles for dereferenceable HTTP URIs, RDF representations, and links between datasets.
A graph-structured representation of entities and relationships, underlying both vendor-proprietary and open enterprise Systems of Intelligence.
A URI-based identity mechanism enabling verifiable, dereferenceable identity for people and agents.
The W3C query language for RDF, proposed as the model-agnostic query interface over an enterprise Knowledge Graph.
The W3C triple data model for representing entities and relationships as subject-predicate-object statements.
The W3C standards stack (RDF, OWL, SPARQL, Linked Data) for machine-readable, interoperable enterprise data.
The open standard bringing Internet connectivity to documents and the entity relationships they contain; the Web is one use case of this.
Components interacting through a shared open standard (HTTP) rather than proprietary bindings, so no single vendor's stack must be adopted to retain interoperability.
Seven steps blending the article's action item with the Linked Data alternative, rendered from the RDF schema:HowTo section.
Catalog which metrics, revenue definitions, approvals and policies are authoritative today, and who owns each decision right.
Give each customer, product, policy, contract, employee and agent a globally unique IRI, following Linked Data principles.
Represent business rules, exceptions and workflows as machine-computable entity relationships informed by ontologies that the enterprise controls, not proprietary Palantir Ontology or Unity Catalog artifacts.
Let any frontier, open, domestic or specialized model query the same standards-based interface.
Give agent traces first-class dereferenceable identifiers and W3C PROV-O provenance chains anchored to verifiable WebIDs.
Treat frontier, open, domestic and small models as interchangeable engines over the same enterprise Knowledge Graph.
Keep governance, auditability and evolution of the SoI under enterprise control, independent of any single vendor's roadmap.
Graph data is derived from the generated RDF entity and relationship model. Node and edge clicks use URIBurner resolver-backed IRIs.
Explore organizations, scenarios, control points, critical perspectives, and article structure.
Graph data embedded from companion RDF at generation time (124 nodes / 264 links). Controls tray is closed by default.