The Last Holdouts Cave: Google and Microsoft Join Apache Ossie, the Open Standard That Teaches AI to Read Your Data

For most of the past year, the two biggest software companies on Earth sat on the sidelines of the most consequential data-standards effort in a decade. This week, that ended within days of each other: Microsoft is now listed on the Apache Ossie ecosystem page, and The Information reports that Google has joined as well — “the latest major software company” to sign on to the industry group working to help AI tools actually understand enterprise data.

On the surface, this is a dry plumbing announcement. Underneath, it is the quiet capitulation of the last two holdouts in the AI data wars — and a signal that the industry has settled its answer to a question that agents keep asking: what does “revenue” mean?

What Apache Ossie actually is

Apache Ossie (Incubating) began life as the Open Semantic Interchange (OSI), launched by Snowflake in September 2025. It entered the Apache Software Foundation’s incubator in mid-2026 and relaunched in July under its new name, with the code and spec on GitHub under the Apache project’s governance.

Its mission is deceptively narrow: standardize semantic model exchange across analytics, AI, and BI platforms. A semantic model is the layer that defines business metrics — datasets, fields, relations, and the logic of measures like “net revenue” or “active users.” Every serious data platform has one; until now, none of them spoke the same language. A metric defined in Power BI’s DAX, a Snowflake Semantic View, and a LookML model in Looker could describe the same business concept three incompatible ways.

The spec defines standard fields for datasets, fields, relations, and metrics, so a semantic model written once can be read and executed by any compliant tool. More than 50 organizations had already joined before this week — Snowflake as initiative lead, plus Databricks, Salesforce, AtScale, ThoughtSpot, dbt, and others, with Nvidia and Oracle counted among the biggest names in the ecosystem.

Why Microsoft’s move is an about-face

Microsoft’s history made this week’s news genuinely surprising. The company spent a year outside the standard while Power BI’s semantic models stayed locked in DAX, built for Power BI and nothing else. Worse, Microsoft had actively moved to block partners from connecting third-party data tools directly to Power BI — a move widely read as ring-fencing Microsoft Fabric at the expense of Snowflake and Databricks, and a front in the emerging AI data wars.

That posture is now dead. The evidence is concrete and public:

• Microsoft is listed on the Apache Ossie ecosystem page alongside Snowflake, Databricks, Salesforce, and AtScale.
• A Microsoft converter sits in the Apache Ossie repository on GitHub, translating Power BI and Fabric semantic models into Ossie format and back.
• The Power BI team published a blog post affirming its “commitment to Apache Ossie,” promising to help customers “define semantic context once and reuse it across ecosystems without duplicating data or business logic.” The company says its current collaboration with Snowflake focuses on cross-platform semantic-layer conversion.

Notably, Microsoft isn’t joining empty-handed — or empty-mouthed. The Power BI team also says it intends to establish DAX as an Ossie-recognized query language and to expand support for ontologies, enabling what it calls “platform-agnostic operational intelligence.” In other words: Microsoft will interoperate, but it wants the standard to speak its language too.

Google’s addition, reported by The Information on September 30, closes the loop. The company that owns Looker and BigQuery — and that has been building its own Universal Semantic Layer on top of them — is now inside the tent as well, though Google has been less vocal about its plans.

The real driver: AI agents can’t open Power BI

The strategic logic behind this reversal isn’t diplomatic warmth. It’s arithmetic.

Few enterprises run on a single vendor anymore. Data sits in Snowflake, Databricks, BigQuery, and Redshift simultaneously. And increasingly, the questions asked of that data come not from Power BI or Excel but from Claude, ChatGPT, and Cursor — AI agents that never open a BI tool at all. A semantic model that only works inside one tool is invisible to every agent that never launches it.

As AtScale’s Chris Lynch put it this week: “The future is going to require a heterogeneous stack that lets any database connect to any LLM or SaaS application.” Microsoft’s customers made this decision long before Microsoft did. Once the industry agrees on how to describe a metric, the holdout’s format is the one no other tool can read — and Microsoft did that math.

The timing also matters. Agentic analytics is shifting from demos to production pipelines in 2026, and every agent that touches enterprise data needs trusted, machine-readable definitions of business metrics. Text-to-SQL without a semantic layer produces confident nonsense; with one, it produces the same number the CFO would get. Ossie is the transport layer for that trust.

What’s still unsolved

The standard has real limits, and they’re worth stating plainly. Ossie gives the industry a common way to write down what a metric means — but someone still has to compute it, identically, every time, for every tool that asks. Two vendors can import the same Ossie model and still produce divergent results if their execution engines differ. That “calculator problem” is exactly where commercial players like AtScale (whose ACE engine computes Ossie-defined metrics on whichever platform the customer runs) and the platforms themselves now compete.

There are also governance questions the incubating project hasn’t fully answered: versioning semantic models across teams, certifying converter fidelity, and reconciling DAX’s bid to become a recognized query language against the SQL-and-semantic-tooling the other members built. Apache incubation is a process, not a finish line.

Why it matters

The significance of this week isn’t that two logos were added to a web page. It’s that the last credible bet on proprietary, walled-garden semantics has been abandoned. When Snowflake launched OSI a year ago, the open standard looked like an alliance of the marginalized. With Microsoft and Google in, the neutral format is now the default — and every vendor still betting on a semantic model that only works inside its own walls has to explain why.

For enterprises, the near-term payoff is cheaper, more reliable AI-driven analysis: agents that read one definition of “gross revenue” instead of reconciling five. For the AI industry, it’s a small but real step toward agents that can be trusted with the numbers. Interoperability stories are rarely exciting. This one redraws the map anyway.

The lesson of Apache Ossie’s breakout year is the same one HTTP, SQL, and USB taught: when a format war drags on long enough, the market picks the door everyone can walk through. This week, the last two giants walked in.