Enterprises Shift AI Workloads to Balance Costs
Databricks' AI Index finds companies increasingly split AI use across multiple model providers to manage spending and output quality.
Updated on Nov. 20, 2026 in Artificial Intelligence

Live Poll
Do you expect the AI tools you use to get more expensive this year?
Databricks published the AI Index assessing enterprise AI use and spend, finding that companies are adopting models from two or more providers and moving workloads between them. IT leaders are seeking to balance costs against the quality of model outputs.
Why it matters
Cost per task depends on both token prices and how much work a model must do to reach an answer, so headline pricing alone does not determine what an enterprise pays. Tracking tokens by app, workflow, agent and user gives companies the telemetry needed to make those trade-offs.
GPT-6 Luna and Claude Opus 5.5 each achieved an 89.1% task success rate, but Claude Opus 5.5 cost $1.62 per task against $0.05 for GPT-6 Luna, a 32-times difference. Gemini 3.8 Flash, with a low token price, cost $11.97 per task, and the figures do not show whether these gaps persist across all task types.
The players
Databricks
Databricks is a data and AI company whose AI Index tracks how enterprises use and spend on artificial intelligence.
OpenAI
OpenAI is an artificial intelligence research company that develops the GPT family of large language models, including GPT-6 Luna.
Anthropic
Anthropic is an AI safety company that develops the Claude family of models, including Claude Opus 5.5.
Microsoft
Microsoft is a global technology company whose corporate vice president and chief technology and AI officer, Tal Saraf, is quoted on enterprise AI strategy.
Gartner
Gartner is a research and advisory firm that tracks technology markets and enterprise IT spending trends.
The details
Enterprise AI spending is highly uneven: a traditional enterprise user spends an average of $2.49 per day on AI, while the 99th percentile user spent $699.74, about 281 times the average. Companies are responding by shifting workloads between providers rather than relying on a single model.
Timeline
2025 is the baseline year against which global AI spend growth of 63% is measured.
2026 is the year global end-user spend on AI models and platforms is projected to reach $64 billion.
Microsoft published a blog post on AI spend return on investment on Thursday.
The Tech Race
The shift toward multi-model enterprise AI strategies reflects a competitive market in which OpenAI, Anthropic and Google offer models at widely different price points for similar task success. Enterprises are now treating model choice as a cost-optimization decision rather than a single vendor commitment.
Users and IT teams choosing AI tools can compare cost per completed task rather than headline token prices, since a cheaper model may require more work to reach the same answer. Tracking usage by app, workflow and user helps teams see where spending is concentrated.
The takeaway
Comparing models on cost per successful task, rather than price per token, gives a clearer picture of what AI actually costs to run. Tracking usage by workflow and user is the practical first step toward managing it.
Further reading
For broader context on how companies are deploying these tools, explore our Artificial Intelligence coverage.
Source note: This article includes information reported by CIO Dive.
Live Poll
Do you expect the AI tools you use to get more expensive this year?



