Semarchy Data Products for Microsoft Fabric: OneLake Semantic Models and Purview Lineage

This is a vendor announcement. Semarchy's claims about what the workload does and how well it works come from Semarchy, not from independent testing. It still matters to Fabric administrators and data platform owners, because it would put a partner workload inside the Fabric portal that builds and maintains semantic models for them. Before that happens, admins will need to plan for capacities, permissions and consent.

What Semarchy announced​

According to Semarchy, the workload publishes Intelligent Data Products from the Semarchy Data Platform for Microsoft Fabric, making mastered data and semantic intelligence available at scale to Microsoft Fabric services including Microsoft Power BI and Fabric data agents, as well as to business analysts and data scientists.

Semarchy says the workload offers a guided experience inside the Fabric portal. It:

• Publishes certified data into Microsoft OneLake and keeps it continuously in sync
• Builds and enriches a Fabric semantic model from the governed business context of each data product, and keeps that model current
• Adds labels and relationships: every entity and attribute lands with a readable label, a clear description, and its relationships to other data
• Pushes governance metadata to Purview: lineage and discoverability are also published to Microsoft Purview, so governance teams can trace every data product back to its source

Semarchy says the output can be used right away by Power BI, Copilot, Fabric Data Agents, Data Warehouse, Data Science and Real-Time Intelligence. The trade publication Compare the Cloud described the same integration as one that automatically synchronises governed business vocabulary, golden records, hierarchies, and data relationships from Semarchy's master data management platform into Fabric. That is still a summary of Semarchy's own positioning, not a hands-on evaluation.

Summary: Semarchy is packaging mastered data plus metadata (labels, descriptions, relationships and lineage) so it arrives in Fabric ready to use, instead of leaving Fabric teams to model it by hand.

Availability and pricing​

The announcement contains two separate dates:

The October Marketplace date applies to the platform, not to the workload. The announcement doesn't give a launch date for the workload, the platform's license price, or which license tiers qualify.

The CLI and the portal workload​

The new workload sits alongside a tool Semarchy already ships. In Semarchy's words, engineering teams that automate everything use the CLI; data platform owners and Fabric administrators use the Workload's guided experience in the Portal. Both deliver the same governed output. The workload follows the Semarchy Fabric CLI announced at Microsoft FabCon EMEA in September 2025.

Semarchy's architecture keeps its own platform as the system of authority for certification, data quality and access control, with Fabric acting as the place where data is consumed. The company says this means there is no second governance plan to manage. That's a design goal, not a guarantee. Fabric has its own workspace roles, OneLake security roles and semantic model permissions, and those still apply to anyone querying the published data.

The partnership isn't new. When Semarchy first announced its Fabric integration in 2025, Database Trends and Applications quoted Dipti Borkar, Microsoft's VP and GM for OneLake and Fabric ISVs: "Together with Semarchy, we're giving our customers additional tools to drive value, automate business insights, and push the boundaries of enterprise data analytics". The tooling story from that announcement still holds. Semarchy's design experience runs in Visual Studio Code, uses GitHub for version control and GitHub Copilot for authoring, deploys on Azure and is governed with Purview.

Why semantic context matters (with caveats)​

Semarchy's pitch is that the semantic model decides whether Fabric's AI features work. The company argues that Fabric users have to manually create a semantic model for their data, and maintain it as schemas change. When dealing with hundreds of entities and thousands of attributes, that work rarely gets finished, and what does get finished drifts from the source.

Microsoft's own documentation partly supports the general point. Microsoft Learn's guidance on Fabric data agents says that when an agent queries a semantic model, it generates DAX using the model's metadata and its Prep for AI configuration. Microsoft also recommends clear, business-friendly names. So well-labeled, well-related metadata does help. What the documentation doesn't show is that Semarchy's workload will make Copilot or agent answers more accurate. Semarchy published no benchmark, customer case study or measured accuracy gain with the announcement.

Microsoft's documentation also sets some limits that apply however good the metadata is:

• Five data sources per agent: a Fabric data agent can use up to five sources in total, across lakehouses, warehouses, Power BI semantic models, KQL databases, ontologies and Microsoft Graph.
• Permissions follow the user: agents read schemas and run SQL, DAX or KQL under the user's Microsoft Entra identity. Read permission on a semantic model is enough to add it to an agent and query it. Write permission is needed to change the model or use Prep for AI.
• No analytics magic: Microsoft says data agents don't currently do advanced analytics, machine learning or causal inference. A question like "what caused our sales spike?" is out of scope.
• Show your work: agents display the intermediate steps and the generated query, which Microsoft positions as a way to check answers.

Better semantics can make an agent less likely to get lost. They don't replace testing, access control design or checking the generated queries.

What Fabric admins should plan for​

Microsoft's generic Workload Hub documentation gives admins a checklist for any partner workload, including this one when it arrives. It isn't a Semarchy-specific onboarding guide.

1. Check who can add workloads. Microsoft Learn says a user needs permission from a Fabric admin to add a workload, and must also be a capacity admin or be allowed to assign capacities to workspaces.
2. Pick capacities on purpose. A workload added to selected capacities is available only in workspaces assigned to those capacities, and only those workspaces can create its items. You can later add or remove capacities under Manage capacities.
3. Expect a consent prompt. The first time someone uses a workload, they must consent to its Microsoft Entra ID application. Tenant admins can consent for the whole organization. Review the requested permissions the way you would for any enterprise app registration.
4. Check the release stage. Microsoft's publishing process for partner workloads has three stages: a preview audience of up to 20 tenants, a public preview listed for all customers with a visible preview label, and general availability after validation. Semarchy hasn't said which stage the workload will launch in. Look for the preview label in the Workload Hub.
5. Watch the unverified-workload settings. Publishers can release workloads to specific tenants before Microsoft certification. Showing those requires two tenant settings: "Capacity admins and contributors can add and remove Partner workloads" and "Users can see and work with unvalidated partner workloads." Keep those settings tightly scoped.

Analysis: a good idea with open questions​

The strongest part of Semarchy's argument is about where modeling work belongs. If an organization has already defined its customers, products and suppliers in an MDM platform, rebuilding those definitions by hand in Fabric duplicates effort and invites drift. Compare the Cloud noted that Semarchy positions the workload as an alternative to hand-built semantic models, which carry a maintenance overhead whenever upstream data products change. For a company with thousands of attributes, automated metadata sync is a real benefit.

The questions are the usual ones. The announcement doesn't cover:

• Sync behavior: how often the data and model refresh, and what happens when an upstream schema change breaks a report
• Scale and capacity: data volumes, supported Fabric SKUs and the capacity cost of continuous sync
• Model ownership: if Semarchy maintains the semantic model, whether it overwrites hand-added measures or conflicts with Prep for AI settings
• Purview detail: which lineage and metadata fields actually reach Purview

The audience is also narrow. The workload is included for licensed Semarchy platform customers, so it helps organizations that already use Semarchy or are evaluating it. For everyone else, it's one more sign that Fabric's partner ecosystem now treats AI-ready semantics as a selling point. The same day's BigDATAwire listings included a ClickHouse Fabric workload and OneLake integration, which suggests the Workload Hub is getting crowded.

Bottom line​

On September 30, 2026, Semarchy announced a Fabric Workload Hub offering that publishes mastered data to OneLake, keeps Fabric semantic models in step with its governed business context, and sends lineage to Purview. It's meant as a portal-based companion to the CLI Semarchy has shipped since March 2026. The workload has no firm launch date, the platform listing is due on Azure Marketplace in October, and Semarchy says the workload carries no extra charge for its licensed customers. The claims about better Copilot and data agent answers are plausible and fit with Microsoft's guidance on metadata, but nobody has shown them independently yet. Fabric admins who want to try it should settle capacity assignment, add-workload permissions, Entra consent and preview status first, then test agent answers against known-good queries before anyone relies on them.