Skip to main content
Enterprise AI, decodedJanuary 1970

May 30, 2026Product Review

Enterprise Intelligence Platforms — Market Review 2026

The question that decides an enterprise intelligence platform is not which has the best analytics — features change every release — but whether it can serve as a shared foundation every tool and agent connects to, or whether it only governs meaning inside its own walls. Judged that way, on four criteria (semantic scope, external connectivity, access-control architecture, and per-tenant configurability), none of the seven platforms here scores well on all four, and that gap is the finding, not an oversight. Two categories emerge. The ontology-native platforms — Palantir Foundry, Stardog, Graphwise — model typed entity relationships an agent can traverse and, in Stardog's case, infer new facts. The analytics platforms with embedded semantic layers — AtScale, dbt, Cube, Looker — deliver real, immediate metric consistency but govern calculations, not relationships, and each adds another private definition as you add tools. The unmet need across both: no platform ships a pre-built industry ontology you can configure and go live on in weeks, which is where the largest value, for buyers and vendors alike, still sits.

12 minKnowledge Graphs & OntologySemantic Layer & Enterprise SemanticsEnterprise Knowledge SystemsInteroperability & StandardsGovernance Risk & Trust
💡 This is the vendor review. It evaluates what the market currently offers against a set of criteria that matter for enterprise-scale AI. For the strategic recommendation on architecture and industry ontology strategy, read the companion piece: Why Enterprise AI Needs One Ontology, Not One Semantic Layer Per Tool.

§ 02

How to Read This Review


This is not a feature comparison. Features change with every product release. What this review evaluates is the architectural position of each platform — specifically whether it can serve as a shared intelligence foundation across an enterprise, or whether it is designed to govern meaning only within its own boundaries.

That distinction matters because enterprise AI has a consistency problem. The more tools and agents an organisation deploys, the more definitions of the same business concepts accumulate in different places. A platform that cannot be connected to a central semantic layer makes this problem worse over time, regardless of how good its own analytics are.

The four criteria used in this review:

  • Semantic scope — Does the platform define metrics only, or does it model typed entity relationships (Customer connects to Order connects to Product) that AI agents can traverse?
  • External connectivity — Can the semantic layer be queried by systems and agents outside the platform, or is it locked inside?
  • Access control architecture — Is access enforced at the semantic layer itself, or delegated to the consuming tool?
  • Tenant and domain configurability — Can business rules and calculations be overridden per organisational unit, operating model, or industry domain without forking the core model?
  • No platform currently on the market scores well on all four. That is the finding, and it is the context for the strategic recommendation in the companion piece.

    § 03

    Category 1: Ontology-Native Platforms


    These platforms treat the enterprise knowledge model as the primary product. BI tools, AI agents, and applications are consumers of it — they do not own the definitions.

    § 04

    Palantir Foundry / AIP


    What it is

    Palantir's Foundry platform is built around its Ontology — a formal model of the enterprise's real-world entities, relationships, and business logic. Objects (Aircraft, Customer, Purchase Order, Field Service Ticket), their properties, the links between them, and the actions that can be taken on them are all defined in one place. Every downstream application — dashboards, workflows, AI agents — reads from this shared model.

    The AIP (AI Platform) layer extends this to agentic use cases. AI agents built on Foundry reason over the Ontology directly, which means they operate from the same business understanding as every other system in the enterprise. The Ontology carries not just what things are but what can be done with them — approvals, escalations, automated actions — which is what makes Palantir agents operationally reliable rather than experimental.

    Strengths

  • Strongest implementation of the central ontology model in the market
  • Ontology carries both semantic and kinetic elements (actions, not just definitions)
  • Access control, governance, and lineage are all ontology-native
  • 200+ data connectors; data stays in place, not copied into Foundry
  • US commercial revenues up 93% YoY in Q2 2025; proven at scale in defence, manufacturing, finance, healthcare
  • Limitations

  • Significant enterprise commitment: deployment in months, not weeks
  • Pricing is enterprise-tier; not appropriate for exploratory or departmental deployments
  • Deep lock-in: the Ontology is Palantir-proprietary, not portable to other platforms
  • Requires forward-deployed engineers or certified implementation partners for initial build
  • Criteria scorecard

    CriterionAssessment
    Semantic scope✅ Full entity-relationship ontology with typed links and actions
    External connectivity✅ API and MCP surfaces for external agents
    Access control architecture✅ Ontology-native, granular, object-level
    Tenant / domain configurability✅ Per-object overrides; multi-tenant supported

    Best fit: Organisations ready to make the ontology the operational backbone of the enterprise, not a reporting layer on top of existing tools.

    § 05

    Stardog Enterprise Knowledge Graph (EKG)


    What it is

    Stardog is an RDF/OWL-native knowledge graph platform with SPARQL reasoning. Its distinguishing architectural feature is that data is never moved — Stardog creates a virtual knowledge graph over existing systems, querying them in place. The ontology connects to your databases, ERP, data warehouse, and other sources wherever they already live.

    AI agents query the knowledge graph and receive governed, relationship-aware answers grounded in the formal ontology. Stardog also supports logical inference — the ability to derive new facts from existing relationships, not just retrieve stored ones. If a regulation implies a classification that is not explicitly stored in the data, Stardog can surface it.

    Strengths

  • No data movement; connects to existing systems via virtual graph
  • Formal logical inference (OWL reasoning) — beyond retrieval into derivation
  • Strongest governance depth in this category for regulated industries
  • MCP integration for AI agent access
  • Free Community Edition available; commercial pricing scales with use
  • Limitations

  • Requires ontology modelling expertise; not self-service
  • SPARQL and OWL have a steeper learning curve than SQL-based tools
  • Smaller ecosystem and partner network than Palantir or the BI-native platforms
  • Longer time-to-value for organisations without existing RDF/OWL investment
  • Criteria scorecard

    CriterionAssessment
    Semantic scope✅ Full formal ontology with OWL reasoning and inference
    External connectivity✅ SPARQL, REST, MCP; external agent access supported
    Access control architecture✅ Semantic-layer-native; fine-grained
    Tenant / domain configurability✅ Supported; requires modelling investment

    Best fit: Regulated industries (finance, healthcare, defence, insurance) with serious governance requirements and willingness to invest in ontology engineering. Strongest for organisations that need logical inference, not just retrieval.

    § 06

    Graphwise (formerly Ontotext GraphDB + PoolParty)


    What it is

    Graphwise is the product of merging two mature semantic web platforms: Ontotext's GraphDB triplestore and Semantic Web Company's PoolParty knowledge management system. It covers both structured data (knowledge graph) and unstructured content (document understanding, taxonomy management).

    Its GraphRAG capability is particularly relevant for AI use cases where answers need to be grounded in both structured knowledge and documents — legal, regulatory, life sciences, publishing. An AI model retrieves semantically grounded answers from the combination rather than hallucinating from context alone.

    Strengths

  • Strong for mixed structured/unstructured AI use cases
  • GraphRAG for document-grounded question answering
  • Mature RDF standards compliance
  • Free GraphDB edition available
  • Good fit for content-heavy industries
  • Limitations

  • Less enterprise-scale deployment track record than Palantir or Stardog
  • Product integration between the two legacy platforms (GraphDB + PoolParty) is still maturing post-merger
  • Narrower applicability outside content and document-heavy domains
  • Criteria scorecard

    CriterionAssessment
    Semantic scope✅ Full ontology; strong on document/knowledge integration
    External connectivity✅ Standards-compliant; open APIs
    Access control architecture⚠️ Supported but less enterprise-hardened than Palantir or Stardog
    Tenant / domain configurability⚠️ Possible; requires configuration

    Best fit: Legal, life sciences, publishing, regulatory compliance — domains where the intelligence layer must span structured data and document content simultaneously.

    § 07

    Category 2: AI Analytics Platforms With Embedded Semantic Layers


    These platforms deliver real and immediate value: consistent metrics, natural language querying, governed access to data. They are the platforms most enterprises are actively deploying today. The limitation — relevant to the strategic argument in the companion piece — is that each carries its own definition of what things mean, creating a cross-tool consistency problem as the number of deployed platforms grows.

    § 08

    AtScale


    What it is

    AtScale is a universal semantic layer — a platform that sits between any data warehouse and any consuming tool, defining metrics once and serving them to BI tools, AI agents, and embedded applications simultaneously. It is warehouse-agnostic: Snowflake, BigQuery, Databricks, Redshift, and others are all supported.

    The GigaOm 2025 Radar named AtScale a leader in semantic layers and metric stores. A Forrester study found 551% ROI with a two-month payback period. Its governance depth — row-level security, column-level masking, object-level security, audit trails — is the strongest in this category. MCP integration means AI agents can query governed metrics without being able to bypass the semantic layer and hit raw tables.

    Strengths

  • Best-in-category governance depth: RLS, column masking, audit trails
  • Vendor-neutral: works across any warehouse and any BI tool
  • MCP server integration for AI agent access to governed metrics
  • Proven enterprise ROI; large install base
  • Excel and MDX support for organisations with legacy Power BI / OLAP workflows
  • Limitations

  • Governs metric calculations, not entity relationships; multi-hop agent reasoning across Customer → Contract → Product is not natively supported
  • No native ABox equivalent — per-tenant KPI formula overrides require schema or deployment workarounds
  • Enterprise pricing; implementation complexity rules it out for departmental deployments
  • Semantic definitions are AtScale-proprietary; some portability via OSI standard but tooling is early
  • Criteria scorecard

    CriterionAssessment
    Semantic scope⚠️ Metrics and dimensions; not typed entity-relationship ontology
    External connectivity✅ Multi-API; MCP for AI agents
    Access control architecture✅ Best in category; semantic-layer-native
    Tenant / domain configurability⚠️ Possible via multi-tenant configuration; not natively ABox-style

    Best fit: Organisations whose primary problem is metric inconsistency across multiple BI tools and AI agents. Not the right choice if the goal is cross-domain entity-relationship reasoning.

    § 09

    dbt Semantic Layer (MetricFlow)


    What it is

    dbt is the dominant analytics engineering platform. Its Semantic Layer, powered by MetricFlow, allows data teams to define metrics in YAML alongside the dbt models that produce them. Any downstream tool — BI platforms, AI agents, APIs — queries those definitions through a consistent interface.

    dbt is also the closest thing the market has to an open standard. The Open Semantic Interchange (OSI), launched in January 2026, uses MetricFlow as its declarative specification. Snowflake, Salesforce, Cube, AtScale, and Databricks are among the 40+ signatories, meaning dbt metric definitions may eventually be portable across platforms.

    Strengths

  • Git-native workflow; metric definitions version-controlled alongside data models
  • Closest to an open standard via OSI / MetricFlow
  • Strong analytics engineering ecosystem and community
  • JDBC and GraphQL API surfaces for LLM and agent integration
  • No additional infrastructure if already on dbt Cloud
  • Limitations

  • Full Semantic Layer API requires dbt Cloud; dbt Core users get MetricFlow only
  • Metric definitions, not domain ontology; governs calculations, not relationships
  • YAML authorship has a learning curve; not drag-and-drop
  • Cross-tenant metric overrides require model branching, not native configuration
  • Criteria scorecard

    CriterionAssessment
    Semantic scope⚠️ Metrics and dimensions; not entity-relationship ontology
    External connectivity✅ JDBC, GraphQL, API; OSI portability in progress
    Access control architecture⚠️ Warehouse-native RBAC; not semantic-layer-native
    Tenant / domain configurability⚠️ Requires model separation; not natively configurable per tenant

    Best fit: Organisations with a strong analytics engineering practice already on dbt, where the goal is metric consistency across BI and AI consumers.

    § 010

    Cube


    What it is

    Cube is a headless semantic layer — it exposes metrics and dimensions through SQL, REST, GraphQL, and MDX simultaneously, making it the strongest option for organisations that need to serve multiple BI tools, AI agents, and embedded analytics applications from one metric definition. The GigaOm 2025 Radar named it an Outperformer.

    Its dedicated AI API endpoint and semantic model agent serve the full data model in a format LLMs can reason over directly — further along in AI readiness than dbt. No dbt Cloud dependency; works across any warehouse, multi-cloud, global deployments.

    Strengths

  • Multi-API surface (SQL, REST, GraphQL, MDX) serves any consuming system
  • Dedicated AI API for LLM and agent integration; most AI-ready in this category
  • No dbt Cloud dependency
  • Strongest for embedded analytics products that need one metric definition across multiple consumer types
  • Outperformer recognition in GigaOm 2025 Radar
  • Limitations

  • Same fundamental scope limitation as dbt: metrics and dimensions, not entity-relationship ontology
  • Row-level security supported but not as enterprise-hardened as AtScale
  • Multi-tenant configuration requires separate data sources per tenant; not native ABox-style overrides
  • Criteria scorecard

    CriterionAssessment
    Semantic scope⚠️ Metrics and dimensions; not entity-relationship ontology
    External connectivity✅ Best in category; multi-API including dedicated AI endpoint
    Access control architecture⚠️ Supported; less deep than AtScale
    Tenant / domain configurability⚠️ Feasible; requires deliberate construction

    Best fit: Embedded analytics products and organisations serving multiple BI tools and AI agents that need one definition to rule all consumers. Strongest AI API readiness in this category.

    § 011

    Looker (Google Cloud / LookML)


    What it is

    Looker pioneered the semantic-layer-first approach in business intelligence. LookML defines all dimensions, measures, access grants, and exploration paths centrally; every query generated by Looker runs through this model. Gemini now queries LookML semantics for conversational analytics.

    It remains the most mature semantic layer inside a BI tool, with a deep track record in large enterprise deployments and a sophisticated access control model (access grants, access filters, user attributes).

    Strengths

  • Most mature semantic layer embedded in a BI platform
  • Sophisticated access control (access grants, user attributes, row-level filtering)
  • Gemini integration for conversational analytics grounded in LookML
  • Deep enterprise track record; large partner ecosystem
  • Limitations

  • Google Cloud lock-in; LookML definitions do not transfer cleanly to other platforms
  • LookML is specialist-driven; derived table sprawl is a documented production problem
  • Pricing ($35K–$150K+/year) and specialist maintenance costs are high
  • Semantic layer is inside Looker; external systems querying it must go through Looker's API
  • No native multi-tenant ontology model
  • Criteria scorecard

    CriterionAssessment
    Semantic scope⚠️ Metrics, dimensions, drill paths; not entity-relationship ontology
    External connectivity⚠️ API access available but locked to Google Cloud ecosystem
    Access control architecture✅ Sophisticated; LookML-native
    Tenant / domain configurability❌ Requires schema or deployment separation; not semantic-layer-native

    Best fit: Large enterprises committed to Google Cloud with dedicated LookML modelling capacity. Not the right choice for multi-cloud or multi-vendor environments.

    § 012

    What the Market Is Missing


    Across both categories, every platform reviewed requires the customer to build the semantic model of their business — whether that is a full ontology or a metric configuration — from scratch at deployment time. There is no platform currently shipping a pre-built, formally specified industry ontology that a customer can configure and go live on in weeks.

    This is the most significant unmet need in the market. The companion piece — Why Enterprise AI Needs One Ontology, Not One Semantic Layer Per Tool — develops the argument for why this is the right strategic move for both vendors and enterprise buyers, and what it would look like in practice.

    § 013

    Selection Guide


    Your situationRecommended path
    Ready to make ontology the enterprise operating layerPalantir Foundry
    Regulated industry, need formal inference, data stays in placeStardog EKG
    Heavy document + structured data use cases (legal, life sciences)Graphwise
    Metric consistency across multiple BI tools is the primary problemAtScale
    Strong analytics engineering practice, already on dbtdbt Semantic Layer
    Embedded analytics product or multi-consumer API requirementCube
    Google Cloud committed, dedicated LookML teamLooker

    No platform on this list scores well on all four criteria. That gap is intentional — it reflects the actual state of the market, not an oversight in the review.

    Correspondence

    New essays to your desk.

    By subscribing you consent to receive our newsletter. Unsubscribe at any time via the link in any email. Privacy Policy.

    Sent only when there is something worth reading. Unsubscribe anytime.