US companies shift AI spend to cheaper models
Reviewed by Helen Jones
Ramp's June 2026 data shows direct US business payments to DeepSeek rising into the top trending SaaS vendors. Open-source inference platforms Fireworks AI, fal AI and DeepInfra also gained spend. OpenAI and Anthropic lost share in the same period. The report marks the first clear monthly evidence of measurable migration away from premium US models toward lower-cost alternatives. Direct DeepSeek usage still carries security and competitive risks according to the analysis. Trend durability remains uncertain. The data covers only Ramp customers making direct vendor payments.
Before this data, most solo founders treated frontier model pricing as fixed infrastructure. They built offers, support flows, and content systems around the assumption that high per-token costs were unavoidable and that quality required the latest closed models. Now the spend numbers show customers voting with actual dollars for cheaper inference. The constraint has moved from model access to model selection and output verification. Founders who still default to premium APIs without testing alternatives are carrying unnecessary cost into every customer interaction and every automated workflow.
Analysis
Stop treating frontier model spend as a fixed cost of doing business. Pick one customer-facing workflow that currently hits a paid API, route it through a local or open-source model this week, and measure both output quality and cost delta before you decide whether the premium tier still justifies itself.
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "US companies shift AI spend to cheaper models", Collab365 Spaces. 1 source referenced.