Yes—enterprise AI has entered a multi-model era, at least among the organizations captured in recent surveys. In Andreessen Horowitz’s 2025 survey of 100 CIOs across 15 industries, 37% said their organizations used five or more models, up from 29% in the prior year. A separate 2025 Cloud Security Alliance report summary hosted by Google Cloud puts the average at 2.6 models per enterprise. The measures and samples differ, so neither figure is a universal count of enterprise adoption.
What does “multi-model” mean for an enterprise?
Here, multi-model means using more than one AI model across an organization. It does not require a particular number of models or a single architecture: a company might select models for different tasks, provide access through cloud services, or run some models in its own environment.
As an Amazon Associate I earn from qualifying purchases.
The clearest recent indicators show adoption, not a census of all businesses. The a16z survey asked 100 CIOs across 15 industries about enterprise generative AI buying and building; its 37% figure refers specifically to respondents reporting five or more models. The CSA figure is an average reported in Google Cloud’s summary of a separate report. Because the populations and measures are not established as comparable, the numbers should be read as corroborating signals rather than combined into one market-wide statistic.
Recommended Free Tools
Attribution matters, too. The CSA page says Google commissioned its report, and a16z is an investor and publisher. OpenAI’s 2025 enterprise report describes analysis of aggregated customer usage data and a survey of 9,000 workers across almost 100 enterprises; it offers a view of OpenAI customers and users, not a survey of the entire multi-provider market.
#1 Best Overall
Why are enterprises using multiple AI models?
Organizations may choose different models because their workloads have different requirements. In a16z’s 2025 findings, matching models to use cases and avoiding lock-in were principal reasons for buying from multiple vendors. Its respondents described differences in performance across tasks such as coding, architecture, writing, and complex question-answering. These are reported observations, not universal rankings of models.
Earlier a16z research also identified performance, model size, and cost as selection factors, along with the desire to access new capabilities quickly. Data control and customization, as well as cloud-hosted versus self-hosted deployment, were part of the discussion. Provider capabilities and commercial arrangements can change, so those earlier observations are useful as evaluation dimensions rather than a current provider comparison.
How should a company choose between AI models?
There is no universally best model or hosting arrangement established by these surveys. Compare options against the organization’s actual workloads, data requirements, infrastructure, and ability to operate them.
- Task performance: Test models on representative internal tasks and assess the results that matter to users. A general benchmark or provider claim alone does not establish that a model will fit a particular workflow.
- Cost and model size: Consider workload-specific cost alongside the capability needed. The most capable model may not be necessary for every task.
- Data control and customization: Determine whether the deployment and adaptation options meet requirements for sensitive information and task-specific behavior.
- Hosting and access: Compare direct provider access, cloud-hosted access, and self-hosting in light of infrastructure, procurement, and operating constraints.
- Operational capacity: Account for the staff and processes needed to evaluate, secure, monitor, and govern each model in use.
What changes when an organization adds models?
More choices can help teams match tools to tasks, but they also increase the work of knowing what is in use and managing it consistently. Google Cloud’s summary of the 2025 CSA report gives an average of 2.6 models per enterprise and says 52% cited sensitive data exposure as their primary AI security risk. The same summary describes governance maturity as a strong predictor of AI readiness, alongside skills gaps and limited understanding of emerging AI-specific risks.
The Partnership on AI’s 2025 report draws on workshops held in December 2024 and February 2025 with more than 20 participating organizations. It identifies responsible-adoption readiness, evaluation and monitoring, compliance, and trust and collaboration across the AI value chain as challenge areas. Its recommendations include formal governance, understanding both official and informal AI use, and educating employees. These are report recommendations, not a claim that one particular governance framework is legally required everywhere.
Quick Recap
Practical operating checks
- Maintain an inventory: Identify models and AI tools used by teams, including informal or unsanctioned use.
- Set oversight and approval: Make clear who can approve a model for a given use and who is accountable for its deployment.
- Evaluate intended behavior: Test each model on the tasks for which it will be used, rather than assuming results transfer from another workflow.
- Monitor deployments: Track performance and issues after rollout, and revisit whether a model remains suitable as capabilities and organizational needs change.
- Review security and compliance: Assess sensitive-data handling and applicable obligations for each deployment context.
- Educate employees: Help staff understand approved tools, risks, and how to raise concerns.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
