• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Tunguz Tweets Anthropic Revenue at $50M Per Megawatt

    Venture capitalist posts on AI companies reselling electricity as cognitive work.

    DF
    ZE
    TT
    5 Sources, 26d ago, first seen 26d ago

    TLDR

    Tomasz Tunguz, founder of Theory Ventures, tweeted that model companies buy electricity by the megawatt and resell it as cognitive work. He stated Anthropic revenue has run as high as $50m per megawatt against $10-15m of compute. The tweet links to an 82-second video alternating between a man in a pink shirt and glasses speaking at a table with on-screen dollar icons.

    Combined views

    113.5K

    5 Sources, first seen 26d ago

    Combined views

    113.5K

    5 Sources, first seen 26d ago

    968 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    968 likes
    52 comments
    313 saves
    42 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    52 comments
    313 saves
    42 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    5 Sources

    @ttunguzModel companies buy electricity by the megawatt and resell it as cognitive work. Anthropic's revenue has run as high as $50m per megawatt, against $10-15m of compute.
    @zephyr_z9Zhipu is monetizing their compute at $8M-10M/MW ($2B ARR & 200MW-250MW of inference compute) Anthropic & OpenAI are monetizing at $40M-50M/MW
    @ShanuMathew93So have seen similar takes like this and this is another topic I see people argue both ways, which cannot definitionally be true. How does one square this commentary with other comments that Anthropic only uses 25-40% of their compute on inference which is how you defend the high monetization rates? Either you're serving a lot more inference which compresses their actual monetization rates or they are using a lot more than marketed compute on training. Both cannot be true. You could conceivably use a narrow and limited amount of compute for a single model training run but I feel like it's foolish to not consider all the R&D/training work that goes into producing outcomes like Astra. It's a result of a lot of training across multiple research projects.
    @DanielleFongRT @ShanuMathew93: So have seen similar takes like this and this is another topic I see people argue both ways, which cannot definitionally…

    5 Sources

    @ttunguzModel companies buy electricity by the megawatt and resell it as cognitive work. Anthropic's revenue has run as high as $50m per megawatt, against $10-15m of compute.
    @zephyr_z9Zhipu is monetizing their compute at $8M-10M/MW ($2B ARR & 200MW-250MW of inference compute) Anthropic & OpenAI are monetizing at $40M-50M/MW
    @ShanuMathew93So have seen similar takes like this and this is another topic I see people argue both ways, which cannot definitionally be true. How does one square this commentary with other comments that Anthropic only uses 25-40% of their compute on inference which is how you defend the high monetization rates? Either you're serving a lot more inference which compresses their actual monetization rates or they are using a lot more than marketed compute on training. Both cannot be true. You could conceivably use a narrow and limited amount of compute for a single model training run but I feel like it's foolish to not consider all the R&D/training work that goes into producing outcomes like Astra. It's a result of a lot of training across multiple research projects.
    @DanielleFongRT @ShanuMathew93: So have seen similar takes like this and this is another topic I see people argue both ways, which cannot definitionally…