OpenRouter: the share of business spending between OpenAI and Anthropic models was roughly even in September, vs. Anthropic commanding a 75%+ share in January
Context & Ripple Effects
Earlier 2026 business-spending indicators favored Anthropic: Ramp found it taking about 73% of first-time AI-tool spend in March, while the share of U.S. businesses paying for its tools rose month over month. The OpenRouter result follows a September week in which OpenAI spending exceeded Anthropic spending for the first time since February 2024.
The two datasets measure different buyer activity, but together they show that Anthropic's early enterprise momentum did not translate into an unbroken lead in routed model spending. Near-parity on OpenRouter makes the competitive comparison less about which vendor is broadly adopted and more about where buyers place marginal workloads.
First-order effects
- OpenAI has erased Anthropic's January advantage in OpenRouter business spending, giving its models comparable standing in a channel where customers can switch workloads between providers.
- Anthropic loses the clearest spending-share signal of its early-year enterprise lead, despite Ramp's March evidence of rising U.S. business adoption for its tools.
Second-order effects
- Enterprise buyers using OpenRouter gain a more balanced supplier set for model-by-model procurement, increasing pressure on both OpenAI and Anthropic to win individual workloads rather than rely on incumbent share.
- OpenRouter becomes more consequential as a comparison layer: spending parity makes its usage data a sharper indicator of workload substitution between the two labs.
Third-order effects
- If routed spending stays balanced, enterprise AI procurement is likely to become more portfolio-based, with model vendors competing for specific tasks and usage budgets instead of a single vendor-wide commitment.
- The divergence between adoption measures and routed-spend share points toward a market in which distribution channels and workload routing shape competitive position alongside customer counts.
The trend: Enterprise AI buying is moving from early vendor adoption toward workload-level, multi-model allocation governed by performance, cost, and routing choice.