The "which data platform" question arrives at some point in every engagement that touches serious volume. In 2026 the shortlist for mid-market buyers is realistically four commercial options — Snowflake, Databricks, Microsoft Fabric, and Google BigQuery — plus a growing self-hosted option built on open formats (Iceberg or Delta, on object storage). Any of these can carry a mid-market estate. The question is which one fits the client, not which one is "best".
The three questions that decide it
1. What is your existing cloud and identity footprint?
If the organisation is already anchored on Azure with Entra ID, Fabric is materially easier to land and integrate, and the licensing is often already negotiated. If it is anchored on AWS or multi-cloud, Snowflake and Databricks are usually cleaner. If it is on Google, BigQuery is the default. The platform that aligns with the existing footprint carries fewer integration and identity costs; that matters more than benchmark performance for most clients.
2. How much of your work is "data engineering in notebooks" vs "SQL analytics"?
Databricks is notebook-first; Snowflake and BigQuery are SQL-first. The choice reflects the team's centre of gravity. A heavy ML team prefers Databricks; a finance-and-analytics-heavy org prefers Snowflake. Fabric genuinely supports both but strongest on the Power BI side. Mid-market clients often pick the wrong one because leadership wanted ML optionality; they then build on SQL for eighteen months.
3. What is your sensitivity to open formats and vendor lock-in?
In 2026, every serious platform supports Iceberg or Delta. Adoption is uneven: Databricks on Delta, Snowflake on Iceberg, Fabric on Delta via OneLake, BigQuery on Iceberg. Clients with a strong preference for open-format-first should lean towards Iceberg and the providers that natively query external Iceberg tables. Clients that value managed simplicity over portability can be more relaxed.
Brief, opinionated take on each
Snowflake
Strengths: SQL ergonomics, storage-compute separation, mature governance, strong marketplace. The default choice for specialty insurance, mid-market finance, professional services. 2026 pricing moves have been modestly friendlier than 2024; still not cheap at high concurrency. Iceberg story is solid.
Databricks
Strengths: unified batch, stream and ML; strongest for heavy ML teams; Unity Catalog has matured. The default choice for larger manufacturing, health data, and anywhere serious ML dominates. Requires more platform engineering discipline than Snowflake; rewards it.
Microsoft Fabric
Strengths: Power BI integration, Microsoft 365 licensing alignment, OneLake's open-by-default posture. 2026 is the year Fabric became genuinely production-credible for mid-market; earlier vintages had gaps. The default choice for Azure-anchored mid-market and most UK public sector.
Google BigQuery
Strengths: serverless ergonomics, strong BI integration, continuously improving ML and vector capability. Less common in the UK mid-market than in the US but a solid choice for Google-anchored orgs.
Self-hosted (Iceberg/Delta on object storage with Trino, DuckDB, or Spark)
Strengths: cost at steady state, portability, no vendor lock-in. Weakness: the operational burden is real; you are the platform team. Appropriate for a minority of mid-market clients with serious platform talent and a long horizon.
What we recommend, by archetype
- Mid-tier accountancy firm (100–5,000 staff): Fabric. Aligned with Microsoft licensing, BI-heavy, modest ML. If M365 is not the standard, Snowflake.
- Specialty insurer (£50m–£1bn GWP): Snowflake or Databricks depending on team. Both have Lloyd's-adjacent precedent.
- Housing association (3k–50k stock): Fabric or Snowflake. Fabric for Azure-anchored, Snowflake for AWS-anchored or multi-cloud.
- Mid-market manufacturer (£50m–£500m turnover, 2–10 sites): Databricks if heavy ML ambition; otherwise Snowflake.
- City council: Fabric in most cases, given the Microsoft estate; exceptions for councils with established AWS or GCP footprints.
- Heavy ML product company: Databricks or self-hosted, depending on platform talent.
What matters more than the platform
After a hundred engagements, we are confident of this: the differences between the platforms matter less than the discipline around them. A well-run Fabric estate out-performs a poorly-run Snowflake estate. Governance, semantic layer, testing, documentation, ownership — these are platform-agnostic and they drive most of the long-term value.
The platform decision matters; just not as much as the discipline decision that follows it.
The migration question
A client on the "wrong" platform is almost never worth migrating on platform grounds alone. The migration cost is high, the opportunity cost higher, and most of the platforms can do most of the job. Migrate when the current platform is genuinely failing a defined workload; do not migrate for architectural preference.