The narrative often pushed is that open-source AI models are eating everything. But enterprise spending data tells a different story. Jesse Zhang, co-founder and CEO of enterprise AI customer service agent platform Decagon, points out that open-source's share of total enterprise large model spending has dropped to 11%. That decline is because most companies are still in early, undefined exploration phases with their AI use cases, so they default to closed-source models. But Zhang emphasizes that once use cases mature, open-source models will take over production environments with their extremely low latency and deep fine-tuning advantages.
https://twitter.com/thejessezhang/status/2074154325933424861
In Decagon's own production environment, 90% of inference calls have already switched to open-weight models. The driving force behind this shift is speed and customization — not cost savings. In customer service scenarios, an 8-second wait for a single conversation completely destroys the product experience. Because top closed-source labs don't offer fine-tuning for their flagship models, and small closed-source models can't be deeply customized, small open-source models fine-tuned for specific tasks have become the only option for high-frequency, real-time interactions.
The future of enterprise AI will see a division of labor: top closed-source labs will continue to lead exploration and discovery in new areas, while open-weight models will increasingly take over actual production for mature businesses. Because model fine-tuning requires massive amounts of data and talent, the migration from closed to open source will be a slow, multi-year process. Both sides will see continued growth during this transition.