news.volyx.in

No, it doesn't cost Anthropic $5k per Claude Code user (martinalderson.com)

480 points by jnord · 175 days ago · 351 comments on HN

Article summary

The article disputes a claim that Anthropic's $200/month Claude Code Max plan can consume $5,000 in compute, arguing that this figure is based on retail API prices rather than actual compute costs. The author estimates that the actual cost is around 10% of the retail price, making the plan profitable for Anthropic. The article suggests that the misconception arises from conflating retail prices with actual costs, and that Anthropic's actual costs are likely much lower. The author also notes that most users do not max out their subscriptions, making the plan even more profitable.

Main themes

  • AI inference costs
  • Retail vs actual costs
  • Anthropic's business model
  • LLM writing style
  • AI industry profitability
  • Compute cost estimation

What commenters say

  • The article's title is sensationalized and refutes a strawman argument that was not actually made.
  • The claim that Anthropic is losing money on inference is based on a misunderstanding of the difference between retail and actual costs.
  • Some commenters believe that the article's writing style is reminiscent of LLM-generated content, which may be a sign of the increasing influence of AI on writing.
  • Others argue that the cost of AI inference is not as high as commonly believed, and that companies like Anthropic can be profitable despite high retail prices.
  • The detection of AI-generated content is a significant challenge, and some companies are working on developing tools to identify and flag such content.
  • The proliferation of LLM-style writing on the internet is seen as a problem by some, who worry that it will lead to a homogenization of writing styles and a loss of unique voices.
  • Some commenters disagree with the idea that LLM-style writing is inherently bad, and argue that it can be clear and effective.
  • The article's estimate of Anthropic's actual costs is seen as plausible by some, who note that the company's ability to prioritize traffic and throttle subscribers can help keep costs under control.