Enterprise AI Footprint Larger Than Model Inventory

Post Views: 1

Enterprise AI ecosystems are significantly larger than previously assumed, with the average organization’s AI footprint three times the size of explicitly listed models, according to a Snyk analysis.

Study Overview

Your enterprise AI footprint is significantly larger than previously assumed, with the average organization’s AI ecosystem three times the size of its explicitly listed models, according to a recent analysis of enterprise AI deployment trends. The study, conducted by Snyk, examined 3,044 enterprise environments and 1.39 million code repositories to assess how organizations are integrating artificial intelligence into their operations.

Key Findings

The research highlights a shift toward complex AI architectures that combine multiple components beyond standalone models. Nearly half of the organizations analyzed did not have any explicitly declared AI models in their codebases, instead leveraging third-party services, packages, and tools to implement AI capabilities. At the same time, 17.2% of enterprises operated extensive fleets of AI models, indicating deep integration across platforms and applications.

Agentic Architectures

A key finding is the growing adoption of agentic architectures, which utilize AI agents, model context protocol (MCP) servers, or both. Over 46.9% of organizations using AI have implemented these frameworks, with more than half deploying full-stack solutions that connect AI agents to MCP infrastructure. This infrastructure enables access to enterprise data, applications, services, and external tools, allowing AI systems to retrieve information, coordinate workflows, and execute actions across environments.

Supporting Technologies

The study also reveals that the AI footprint extends beyond models to include supporting technologies such as frameworks, retrieval systems, vector databases, datasets, and other tools. When these elements are factored in, the average AI footprint exceeds three times the size of model inventories. This complexity introduces additional dependencies, integration points, and governance challenges that organizations must address.

Model Providers and Deployment

In terms of model providers, OpenAI remains the most widely used, though competitors like Anthropic have gained traction. The top four providers accounted for approximately 71% of identifiable model occurrences, reflecting a diverse but concentrated vendor landscape. Proprietary models dominated deployments, making up 63.8% of all models, while open-source models represented 32.5%. Proprietary systems are often used for advanced reasoning and autonomous tasks, whereas open-source models are frequently deployed for embeddings, retrieval, and supporting workloads.

Third-Party Dependencies

Third-party software plays a critical role in enterprise AI, with 77.4% of AI packages and tools sourced externally. This reliance on external dependencies introduces security, governance, and supply chain risks that require careful management. The report emphasizes the importance of visibility into these ecosystems, as organizations struggle to trace the lineage of AI models and understand how datasets influence model behavior.

Transparency and Governance

A significant challenge identified is the lack of transparency in model training and fine-tuning. Approximately 50% of organizations using AI models could not link their systems to the datasets used for training, complicating efforts to investigate incidents, ensure compliance, and audit model outputs. This gap underscores the need for stronger governance frameworks and clearer documentation of AI workflows.

Industry and Geographic Adoption

AI adoption varies across industries, with media and entertainment companies leading in the concentration of AI components per organization. Retail, consumer goods, and education sectors also showed high levels of integration. Technology and IT firms, however, deployed the largest overall volume of AI systems. Sectors focused on content creation, customer experiences, and business automation reported the highest adoption rates, often incorporating agents, orchestration frameworks, and supporting infrastructure to enable cross-system functionality. Geographically, North America and Europe exhibit similar AI architecture trends, with both regions increasing their use of AI agents and MCP infrastructure. North American organizations tend to deploy AI at a larger scale, though the underlying technologies and design patterns remain consistent across regions.

Conclusion

The report underscores the evolving nature of AI in enterprise environments, where systems are becoming more capable and interconnected. While organizations often rely on established proprietary models for production workloads, the growing use of open-source alternatives for specialized tasks is narrowing the gap between commercial and open-source solutions. This shift provides enterprises with more flexibility but also requires robust strategies for managing the expanding AI ecosystem.