Databricks Releases AI Index With Key Enterprise Trends
Databricks' AI Index reveals open-weight models now account for 30.9% of enterprise tokens, and 72% of companies use multiple model providers. This shift may impact enterprise buyers' expectations and pricing strategies.
Original reporting · saastr.com
Original: “5 Interesting Learnings from the Databricks AI Index: Open-Weight Jumps to 31% of Enterprise Tokens, 72% of Companies Now Use Multiple Model Providers, and a 281x Gap Between…”
Read “5 Interesting Learnings from the Databricks AI Index: Open-Weight Jumps to 31% of…” at SaaStr →
How this was written. Awesome B2B wrote this article with a language model from SaaStr’s reporting (model: @cf/meta/llama-3.3-70b-instruct-fp8-fast); it describes that reporting rather than adding to it. The original, linked above, is the full story.
Key points from SaaStr’s report
- Open-weight models now account for 30.9% of enterprise tokens, up from 7.3% in July.
- 72% of companies use multiple model providers, treating models like cloud regions.
- Anthropic leads the proprietary side, with its overall share rising from 57.6% to 68.8%.
- Measuring cost per completed task is crucial, as model efficiency and cost-effectiveness can vary significantly.
- Pricing strategies may need to adapt to the large disparity between median and P99 users' spending
The Databricks AI Index, based on token traffic from over 10,000 companies, provides insights into enterprise AI adoption, according to SaaStr. One key finding is the rapid growth of open-weight models, which now account for 30.9% of enterprise tokens, up from 7.3% in July.
This shift may be driven by companies seeking to cut costs on high-volume work, such as classification, extraction, and summarization. Open-weight leaders, mostly Chinese labs, are gaining traction, with GLM 5.2 accounting for about 40% of open-weight tokens on the gateway.
Another significant trend is the decline of single-provider companies, from 64% to 28% in one quarter. Enterprises now treat models like cloud regions, using multiple providers and routing workloads accordingly. This change may make it difficult for companies to maintain a competitive edge by relying on a single model provider.
Anthropic still leads the proprietary side, with its overall share rising from 57.6% to 68.8%. The launch of Opus 5.5 contributed to this growth, with enterprises quickly adopting the new model. However, the best model is not always the most used, as companies prioritize cost and efficiency.
The index also highlights the importance of measuring cost per completed task, rather than price per million tokens. A model's efficiency and cost-effectiveness can vary significantly, impacting gross margin. For example, Gemini 3.8 Flash has a low token price but costs $11.97 per task, making it one of the most expensive models in the benchmark.
The median user spends $2.49 a week, while the P99 user spends $699.74, a 281x difference. This disparity may lead to changes in pricing strategies, with companies adopting usage caps, credits, or tiers to track consumption. The budget for AI is real and may come from beyond software budgets, with CFOs likely to take notice of the 'AI bill' in the future.
Additional findings from the index include the potential for better models to cost less, the push for open-weight models due to frontier guardrails, and the growing importance of data retention as a selling point. As the AI landscape continues to evolve, enterprises and providers must adapt to these changing trends and priorities.
Questions and answers
What is the Databricks AI Index based on?
The Databricks AI Index is based on token traffic from over 10,000 companies that route their AI through Databricks' Unity Gateway.
Why are companies adopting open-weight models?
Companies are adopting open-weight models to cut costs on high-volume work, such as classification, extraction, and summarization.
What is the significance of the large disparity between median and P99 users' spending?
The large disparity may lead to changes in pricing strategies, with companies adopting usage caps, credits, or tiers to track consumption, as the budget for AI is real and may come from beyond software budgets.

