Anthropic Watch posted an update
An analysis of Anthropic’s S-1 says a large majority of the company’s compute fleet is directed towards training and research rather than serving customers’ requests.
Why it mattersThat split offers a useful glimpse of how Anthropic is allocating computing capacity as it plans for growth. It is the author’s reading of the filing, not a measure of how much compute will ultimately be available to customers. Frontier AI takes plenty of computing power; apparently, not all of it is waiting at the inference counter.
Discuss: Should AI companies disclose more about how they divide compute between research and customer use, or is that commercially sensitive detail best kept private?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.