OpenAI and Anthropic each draw 80% of their enterprise revenue from the top 1% of their customers, according to data from Ramp. Lead economist Ara Kharazian said on LinkedIn that no other software category the company tracks is this heavily concentrated.
Two customers account for much of Anthropic’s share, with Cursor and GitHub Copilot together driving roughly $1.2B of the $5B the company reached last year. GitHub, of course, is owned by Microsoft, which has invested $13B into OpenAI since inception.
OpenAI hasn’t said what any single customer contributes, though it reported that its API has served over 10B tokens apiece to more than 9,000 organizations, with close to 200 crossing 1 trillion.
This revenue consolidation could prove to be a major problem for OpenAI and Anthropic in the future. Those same heavy-spenders that make up 80% of their revenue are the same companies that are most likely to explore cost-cutting measures in the future, such as utilizing open-source models or bringing certain tasks in-house.
A few weeks ago I reported that Shopify replaced a frontier model it was using to answer merchant questions about their store with a much smaller internal model it retrains every day on its own production failures. The result was a system Shopify estimates will cost around $1M a year instead of $27M (a 96% savings), while running 38% faster and needing 14% fewer GPUs. Now imagine when every enterprise company starts doing that.
Just last week, the New York Times published an exposé entitled Corporate America Is Getting Hooked on Open-Source A.I., which shared stories of how enterprises like AT&T, Deloitte, and Airbnb are increasingly using cheap, freely available AI models over expensive ones from Anthropic and OpenAI.
Meanwhile, OpenAI is cutting off billion-dollar customers as if there are countless others lining up around the corner. These companies better be careful with their revenue and expense management, or those pricey data centers might turn into Halloween Superstores before they know it.
