What Everyone Gets Wrong About Enterprise Open Source AI

The share of open source in current enterprise LLM spending has dropped from 19% a year ago to 11%. Contrary to the common belief that this reflects open source being pushed out, it actually suggests that the way enterprises use AI is changing.

Looking at Decagon's case, they currently run about 90% of their workloads on open source models instead of OpenAI or Anthropic. This is driven less by cost savings and more by the need for Latency and Fine-tuning. In real-world environments like customer service, small, fast models optimized for specific tasks are essential rather than general-purpose giant models.

Shifts in enterprise AI market share are determined by Use case maturity.

• Early adoption stage: Highly intelligent Frontier models are used to define the shape of the problem, and during this process, spending on closed models surges.
• Maturity stage: Once the distribution of input data and failure cases are understood, models are Distilled and Fine-tuned into small models optimized for specific tasks, shifting toward open source.

Ultimately, the current decline in share is a temporary phenomenon caused by an explosive increase in new AI use cases. All frontier-model-based prototypes will eventually go through a maturity phase and transition into open-source-based production models, with Frontier labs handling Discovery and Open source handling Production.