“During the first half of this year, I thought that Anthropic and OpenAI were benefiting a lot from token maxing. That was where enterprises told their employees to use AI and to hell with the cost. But that ended this summer when costs got out of control and companies now had the option to switch to cheaper open weight models. So I assumed that once token maxing ended, growth would slow…”

…really counts. I also show the company's compute and infrastructure expense during the same periods. And I then calculate Anthropic's efficiency ratio, which I define as its revenue divided by the compute expense. And I also show the sequential percentage change in quarterly revenue. Now, for those of you who are only listening, I think a clear picture emerges, and it is what I have feared. During the first half of this year, I thought that Anthropic and OpenAI were benefiting a lot from token maxing. That was where enterprises told their employees to use AI and to hell with the cost. But that ended this summer when costs got out of control and companies now had the option to switch to cheaper open weight models. So I assumed that once token maxing ended, growth would slow. Still, there would be growth, but slower. And I think that is what these numbers seem to show. If we go back to the fourth quarter of last year, just before token maxing, Anthropic's quarterly revenue grew sequentially by 43%. Its revenue divided by compute expense was 0.69 times. Then token maxing took off, and so did Anthropic. Q126 and Q226 sequential revenue growth jumped to 135% and then…