Open models are taking a bigger bite of the AI traffic passing through Databricks’ Unity Gateway. Announcing its AI Index on November 16, 2026, the company put models it calls open source at roughly 30% of tokens in a gateway sample spanning more than 10,000 companies. Source Source Source
According to @alighodsi’s post, 75% of enterprises in the sample use two or more model providers, and 50% use three or more. Six months earlier, most used just one, the post said. Not exactly a one-brand household. Source
Those percentages overlap: enterprises using at least three providers also belong to the two-or-more group. And the figures count providers, not individual models. Source
A fast rise, with a timeline wrinkle
@alighodsi describes growth from a single-digit token share to more than 30% over six months. @pwendell’s thread instead says nearly zero to 30% over five months. Both put the share at roughly 30%, but the exact starting point and comparison period remain unclear. Source Source
In @pwendell’s anecdotal account, enterprises use those models mostly to reduce costs on high-volume workloads, including large batch jobs and coding. The post doesn’t report measured savings or establish why token share rose. A token counter isn’t a savings receipt. Source
Gateway traffic, not the whole market
Databricks describes the index as covering companies ranging from digital natives to Fortune 500 and Global 2000 businesses. The posts don’t establish that this gateway sample represents enterprise AI use as a whole. Source Source
The roughly-30% figure measures token share within that sample, not the percentage of enterprises adopting open models. The available post text doesn’t define “open source” or explain how tokens and providers were counted. Source Source Source
Comments