Gartner forecasts open-weight models will handle 90% of enterprise AI agent steps by 2030

Gartner analysts Mark Petty and Raymond Paquet predict that open-weight models will power 90% of enterprise AI agent workflow steps by 2030, a shift driven primarily by cost advantages and control benefits. The forecast appears in a press coverage piece on the prediction link.

Petty and Paquet note that open models can match closed-model quality on classification, tool calling, data validation, and reconciliation tasks, while costing only a quarter of the cheapest closed rival and as little as a twentieth of a frontier model. After accounting for redundancy and operational inefficiencies, they estimate practical savings of roughly five to six times.

The analysts argue that enterprises will favor domain-specific, fine-tuned open models for high-frequency, stable tasks that run millions of times per month, allowing small per-step savings to compound. Self-hosting also preserves control, reducing reliance on external vendors.

Our read: The projection hinges on the assumption that cost parity will be achieved without sacrificing quality, and that organizations are willing to absorb the operational overhead of self-hosting. If those conditions hold, the shift toward orchestration could reshape vendor dynamics in the enterprise AI market.