As every major data platform rushes to build its own semantic layer, customers are trading one form of lock-in for another. The open data era solved storage. The next battle is about who owns the business meaning.
Apache Iceberg and open catalogs cracked the storage layer wide open. Bidirectional Iceberg access between Snowflake and Microsoft Fabric reached GA in January 2026. But below the data sits an entirely different layer — business meaning — and that one is fragmenting.
The physical data format: files, partitions, table metadata, and the query catalog. Apache Iceberg provides an open format; REST catalog APIs make it engine-agnostic.
The business meaning layer: metric definitions, dimensions, business rules, data contracts, and governance policies. Each platform writes this in a different proprietary YAML format.
A customer who defines "Net Revenue" inside Databricks Unity Catalog cannot take that definition to Snowflake without rebuilding it from scratch. A customer who leaves Informatica PowerCenter leaves behind not just ETL code but years of embedded business meaning — transformation rules, glossary terms, data lineage — that no migration tool currently carries over.
The Open Semantic Interchange (OSI) started as a Snowflake-led initiative to define a vendor-neutral YAML format for business metrics. It is an interchange format, not a new semantic layer product. Define once, let any engine read it.
Snowflake leads formation. Launch participants: dbt Labs, Salesforce, BlackRock, RelationalAI. Spec work begins under Apache 2.0.
First complete spec covering datasets, metrics, dimensions, relationships, and contexts. Published to GitHub under Apache 2.0. Four reference converters committed: dbt MetricFlow, GoodData, Salesforce, Apache Polaris.
Project moves to neutral governance under the Apache Software Foundation. Now backed by 60+ organizations. De-facto standard status targeted for 2027.
Dosi announced as the first product to execute Apache Ossie YAML to SQL across 15+ warehouse dialects. Cube and Tableau integrations still on roadmap (Tableau: late 2026).
References to underlying tables or views. Platform-agnostic pointer to physical data.
Computed business measures (revenue, churn, MAU) with SQL expression logic.
Attributes for slicing and filtering (region, product, time grain).
Join paths between entities — how datasets relate to each other semantically.
Named query scopes that bundle a metric set for a specific domain or persona.
Machine-readable format that AI agents can query to understand business terminology.
Every major platform defines "a metric" differently. This is where semantic portability breaks down today.
| Platform | Semantic Object | Format | Query Method | Open Source? | Ossie Support | Portability Risk |
|---|---|---|---|---|---|---|
| Databricks | Unity Catalog Metric Views | YAML — version, source, dimensions, measures |
MEASURE() function in SQL |
Yes (2026) | Roadmap | Medium |
| Snowflake | Semantic Views + Cortex Analyst YAML | YAML — different schema from Databricks for same constructs | Cortex Analyst NL queries + SQL views | No | Roadmap | High |
| Microsoft Fabric | Power BI Semantic Models | DAX + tabular model (TMDL YAML since 2025) | MDX / DAX / XMLA | Partial | Not yet | Very High |
| dbt | MetricFlow Semantic Layer | YAML — semantic_models, metrics blocks |
MetricFlow CLI / dbt Semantic Layer API | Yes (MetricFlow) | Converter merged | Low |
| Teradata (legacy) | Business Metadata / Viewpoint | Proprietary — no public spec | SQL + BTEQ scripts | No | None | Very High |
| Informatica (legacy) | Business Glossary + Axon Catalog | Proprietary catalog XML / REST API | EDC / Axon UI | No | None | Very High |
When customers migrate from legacy platforms, migration tools move SQL and pipelines. Nobody moves the meaning. Here's where semantics actually lives in each legacy system.
Semantic portability is a services-heavy problem. The tooling is either missing, immature, or requires significant configuration. This is a window for any IT services firm with migration, data platform, and governance practice depth.
Today, migration projects move SQL and pipelines but abandon business meaning. A services partner with automated code migration capability should extend it: extract business semantics from legacy repositories (Informatica Business Glossary, Teradata metadata, SAS libraries) and emit Apache Ossie YAML as a parallel migration deliverable — not a post-migration cleanup item.
Apache Ossie targets de-facto standard status in 2027. The implementation playbook doesn't exist yet — no reference architecture, no governance model, no migration patterns. The services partner who builds this first becomes the go-to for every enterprise migration that follows, and earns preferred partner status with both Snowflake and Databricks.
For customers running Databricks + Snowflake in parallel, the services play is a neutral semantic orchestration layer — federated metric definitions maintained in dbt MetricFlow or an Ossie-compatible format, compiled to each platform's native dialect. Prevents metric drift while the Ossie ecosystem matures into native platform support.
BCBS 239, IFRS 9, FRTB, and Solvency II all require auditable, platform-agnostic definitions of risk and financial metrics. A bank migrating from Oracle OBIEE to a modern platform cannot lose the certified definitions of "RWA", "PD", or "EAD." This reframes semantic portability as a compliance requirement — a far stronger executive conversation than a data engineering efficiency argument.
| Signal | Current State (Oct 2026) | 2027 Trajectory | Risk to Customers |
|---|---|---|---|
| Apache Ossie adoption | Spec stable, 4 converters merged, Apache Incubator phase | Databricks + Snowflake native import/export converters expected | Low — open standard, neutral governance |
| Platform semantic investment | Each platform doubling down on own format (UC metric views, Cortex Analyst) | Competitive pressure from Ossie will force openness or fragmentation deepens | High — currently incompatible and accelerating |
| Legacy semantic extraction | No automated tool extracts business meaning from Teradata/Informatica/SAS to Ossie | Opportunity window: 12–24 months before someone builds this at scale | Very High — semantics lost silently on every legacy migration today |
| AI agents + semantic layer | AI agents need machine-readable semantics to query correctly; all platforms building this | Ossie becomes the lingua franca for AI agents querying enterprise data | Lock-in risk multiplies if AI tooling tied to platform-specific semantic formats |
| Data contracts portability | dbt Data Contracts, Snowflake Horizon, Databricks UC — no interop | Open Data Contract Standard (ODCS) gaining traction separately from Ossie | Medium — emerging but not yet critical path |