Revenue Lags at AI Evaluation Startups

Startups whose products evaluate artificial intelligence to make sure it isn’t making things up or delivering unsafe results, like Patronus and Braintrust, have caught the attention of investors. But previously undisclosed sales data from these and other startups selling evaluation software show corporate customers have been relatively slow to buy such tools. 

Take Weights & Biases, an eight-year-old startup that CoreWeave agreed to acquire for more than $1 billion last month. A year ago, the startup, which helps developers build AI applications and models, started selling a product called Weave that helps startups run, test and evaluate AI models, including to detect hallucinations. 

By December, Weights & Biases was generating $50 million annual recurring revenue, translating to a little over $4 million a month in revenue overall. But only 2% of the annual revenue pace came from Weave, according to two people with direct knowledge of its financials.

Revenue at other startups that specialize in AI model-evaluation software has also been modest.