The post pulls specific serving-cost and compute figures from Anthropic's S-1 filing, showing how inference economics changed as revenue grew.
The era of efficiency
- Inference: serving got much cheaper - fell from 88% of revenue to 15% in a year, with revenue up 14× and the serving bill up only 2.3×. Serving is only about a quarter of their roughly $8.9B compute bill. The filing names open-weight models as a pricing risk 👀👀
- Training: Claude helping build the next Claude (everyone is doing it now), and running on four kinds of chips. Compute outside serving, mostly training and research, grew about 5×.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO. Read more: https://www.anthropic.com/news/s1-sec Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.View quoted post on X
