A published report that frontier artificial-intelligence lab Anthropic was profitable for a second quarter in a row contains a very important caveat.

The report, from the Financial Times citing multiple people with knowledge of the matter, said Anthropic told investors that it will profit on a measure called adjusted operating income.

According to the report, adjusted operating income excludes stock-based compensation, which is not unusual for technology companies. However, the report also quotes its gross margins — above 80% — exclude revenue shared with distribution partners, as well as the cost of training its models.

A message to Anthropic on its accounting definitions wasn’t returned.

Training costs can be massive. At SpaceX, which admittedly is behind OpenAI and Anthropic in attracting users, the AI segment’s research-and-development costs of $2.03 billion in the second quarter outpaced its $1.47 billion in revenue. SpaceX says the AI segment’s R&D expenses “mainly relate to the training of Grok, our leading frontier model, development, build, and testing of our next-generation AI-enabled products and data center costs to train AI-enabled products.”

The partner revenue split is also significant, though likely not as much as training costs. The Information reported earlier this year that Anthropic’s revenue split to partners including Amazon ran at about 10% of sales, though that was during a period when the Claude large-language-model maker was bringing in less in revenue.

Anthropic’s economics will become clearer once the company publishes its initial public offering prospectus.

Anthropic CEO Dario Amodei over the weekend called for a slowdown in the pace of bringing out new AI models.

A futures contract tied to Anthopic’s valuation on the Hyperliquid exchange was implying a $2.15 trillion valuation when it goes public.