Anthropic Takes 60.5% of Ramp AI Spend Despite Higher Prices

What happened

On 6 Nov 2026, Ramp published corporate card and billing data tracking enterprise AI token expenditure through the week ending October 23. Anthropic's share of business AI token spend rose to 60.5% from 51.3% four weeks earlier, while OpenAI's share fell to 35.1% from 44.2%.

Over the same four-week window, OpenAI's blended price per million tokens fell 42.4% to $0.46, while Anthropic's blended price fell 15.7% to $0.61. Businesses in Ramp's sample used 37.1% more AI tokens in the week ending October 23 than a month earlier.

Separately, OpenRouter reported on 7 Nov 2026 that monthly platform volume reached 790 trillion tokens in October, with autonomous agent workloads consuming 76% of all tokens, up from 8% a year earlier.

What the chart shows

The grid divides corporate AI token spending into a hundred squares. Anthropic captures 60.5% of total dollar spend across Ramp's business sample, OpenAI captures 35.1%, and all other providers account for the remaining 4.4%. Despite charging a higher blended price of $0.61 per million tokens versus OpenAI's $0.46, Anthropic gained 9.2% of wallet share in a single month.

The underlying elasticity is visible in Anthropic's amended IPO filing figures cited in the report. Price reductions and contractual discounts reduced third-quarter revenue growth by about $600 million compared with the prior quarter, yet surging token consumption more than offset the discount, lifting quarterly revenue by $5.4 billion to $17 billion.

Two structural caveats apply to the spending data. First, Ramp's sample reflects corporate card and invoice payments from U.S. businesses rather than consumer subscriptions. Second, Anthropic's revenue remains concentrated at the top end, with about 21% of third-quarter revenue coming from its top five customers, while computing costs rose to 58% of revenue.

What it means for founders

Cutting token prices by 42.4% did not protect OpenAI's share of enterprise billing when developer workflows standardized on coding and agentic harnesses that favor Claude. Once an engineering team embeds a model into its automated agent loops and IDE tooling, switching requires re-validating prompts, tool-call schemas, and evaluation suites.

OpenRouter's disclosure that agents now account for 76% of all platform tokens explains why dollar share follows workflow lock-in rather than headline per-token discounts. Human chat sessions are price-sensitive and easy to reroute; multi-step coding and enterprise search agents run thousands of background calls where reliability dominates unit token cost.

For application founders, this means model providers cannot buy away your customers with raw API price cuts if you own the orchestration harness and domain verification layer. However, because 58% of Anthropic's revenue goes back into compute, foundation labs will keep pushing upward into application workflows to defend margins.

What would change our mind

We would change our mind if OpenAI's $0.46 per million token pricing triggers a migration of automated agent workloads back from Anthropic over the next two months, pushing OpenAI back above 44.2% of Ramp business spend. A sustained reversal in corporate billing share would show that price sensitivity eventually overrides developer harness inertia once API discounts cross a threshold.

We would also revise our view if open-weight routing layers compress Anthropic's enterprise wallet share below 51.3% without any change in coding benchmark leadership, proving that automated routers can arbitrage model prices without workflow friction. Until enterprise buyers decouple their agent orchestration state from model-specific tool conventions, workflow integration will continue to outweigh headline token discounts across corporate billing ledgers.

OpenAI cut blended token prices 42.4% to $0.46, yet Anthropic grew to 60.5% of corporate token spend because agent harnesses lock in workflows.

What we would back: Vertical agent harnesses and verification layers that own the workflow state while routing commodity token compute underneath.