Microsoft joins Apache Ossie, an open metrics standard

At FabCon Europe in Barcelona, Microsoft confirmed Apache Ossie and published a Power BI converter.

In late September 2026, at FabCon Europe in Barcelona, Microsoft stated its commitment to Apache Ossie (incubating). Ossie is an open standard for semantic metadata and business metrics: a way to write down, in text, what a company means by revenue, churn, or margin, so other tools can read the same definition. Arun Ulag, executive vice president of Azure Data, wrote on the Azure Blog on 28 September: “We’re also affirming our commitment to Apache Ossie (incubating)”. On 30 September Snowflake, which started the original effort, said Microsoft engineers had joined the project community.

People who run Microsoft Copilot while the warehouse sits somewhere else already know the effect. Without a shared dictionary, an agent mixes billed revenue, contracted revenue, and cash. The announcement only goes part of the way. Microsoft joined the standard and published a converter. It did not publish proof that agents now read the same indicator.

Quick summary

• On 28 September 2026 Microsoft confirmed Apache Ossie on the Azure Blog, during FabCon Europe in Barcelona.
• The Apache GitHub repository now has a two-way converter between Power BI and Fabric models and Ossie YAML, with documented losses.
• Snowflake says the spec already introduces DAX, Power BI’s formula language, alongside ANSI SQL. There is no full parity.
• The Information described the step as a walk-back from a walled garden. Microsoft did not.
• Treating AI agents as deterministic readers of the same KPI remains RUMOR.

First: what Apache Ossie is

Ossie, previously called Open Semantic Interchange (OSI), describes datasets, fields, metrics, and relationships in YAML. YAML is structured text, with names and values, easy to put in version control. A company writes down once what it means by revenue or by churn, the share of customers who leave, and other tools can read the same card.

Think of a shared glossary. The real mechanism is a specification file, not a calculation engine. The file does not run the math. Each platform still computes inside its own engine. That is where the glossary comparison stops: the same definition does not guarantee the same number if the engines differ.

Snowflake, Salesforce, dbt Labs, BlackRock, and RelationalAI presented OSI on 23 September 2025. Microsoft was not on the launch release. On 10 July 2026 the project entered the Apache Incubator, still not a top-level Apache Software Foundation project, under the name Apache Ossie (Incubating). The ecosystem page lists Microsoft, Snowflake, Databricks, Salesforce, dbt Labs, and dozens of other logos.

What Microsoft shipped

Microsoft engineering published the converters/microsoft tool in the Apache GitHub repository. It translates both ways between Power BI and Fabric semantic models, in TMSL (model.bim) and TMDL, the model languages, and Ossie YAML. The Python package ossie_microsoft runs on the command line. Full TMDL conversion optionally needs TOM, Microsoft’s proprietary software.

Microsoft says it is “helping establish DAX” in the spec. DAX, Data Analysis Expressions, is Power BI’s formula language. Snowflake, on 30 September 2026, says recent spec updates already introduce DAX, alongside ANSI SQL. The converter maps measures to the DAX dialect. There is no full parity with SQL.

In a Fabric Community post, Microsoft’s Christian Wade says Power BI semantic models have more than 38 million monthly active users and are used by 94% of Fortune 500 companies. The figures come from Microsoft, in the FabCon Europe 2026 week, with no independent audit. They do not measure Ossie adoption.

What The Information said

The Information, in public posts on 29 and 30 September 2026 and in the paywalled article “Microsoft Tears Down Its AI Data Wall”, called the step an “about-face from Microsoft’s earlier effort to protect Fabric from rivals”. The full article sits behind a paywall; this newsroom did not read it in full.

In official posts Microsoft talks about interoperability customers asked for and about agentic experiences, software that calls tools on its own. It does not admit walking back a walled garden. That reading is press analysis, not a Microsoft statement. Microsoft and Snowflake are announcing a standard both of them use. Read the releases with that discount.

Limits, and what is still unknown

The converter README documents losses: anything that does not fit simple SQL aggregations is emitted as BLANK(). The offline converter does not parse or validate DAX. Only validate_with_engine on Fabric or Analysis Services compiles the formulas. There is no public date for a native export/import Ossie button in Power BI Desktop or the Fabric service.

The Ossie site lists “Hallucinations / Unreliable AI grounding” as the problem the standard tries to attack. That is the project’s thesis, not proof Ossie already fixes it. Treating AI agents as deterministic readers of the same KPI, the indicator a company chose to track, remains RUMOR. Microsoft has not measured that interoperability. This newsroom did not run the converter.

What is in the Apache repository is a public commitment and a model translator with documented gaps. That matters if you already keep semantics in Power BI and want to carry the definition to another engine. It does not replace a test on your own model. It also does not settle whether a wall came down.

Sources

Official

• Microsoft Azure Blog, Arun Ulag, 28 September 2026
• Microsoft Fabric Community, Christian Wade
• Snowflake Engineering Blog, Josh Klahr, 30 September 2026
• Apache Ossie, official site
• Apache Ossie, ecosystem
• Apache Ossie enters the Incubator, 10 July 2026
• GitHub, converters/microsoft
• GitHub, Microsoft converter README
• Snowflake, OSI announcement, 23 September 2025

Press

• The Information on X, 30 September 2026
• The Information, Microsoft Tears Down Its AI Data Wall