Enterprise AI strategies are shifting toward multi-model architectures as companies increasingly adopt open-weight models to improve cost efficiency, data governance, and operational flexibility over single-provider frontier solutions.
Key Points
- Databricks reports that the average number of unique models used per customer on its Unity Gateway increased from 4.2 to 7.1 since January 2026.
- Open-weight models now account for approximately 40% of token usage on the Unity Gateway platform, driven by the release of high-performing models like GLM 5.2.
- Enterprises are moving away from single-provider dependencies to avoid issues with restrictive guardrails, evolving data retention policies, and fluctuating model rankings.
- Databricks raised $3 billion in July 2026, with CEO Ali Ghodsi citing the need to secure massive GPU capacity to host and serve open-weight models.
- Unity Gateway allows organizations to switch between frontier and open-weight models via configuration changes rather than code, maintaining consistent governance and auditing.