Eric Siu spent 3.5 billion tokens in a month and paid $5,600 to Anthropic. He argues that engineering teams need to shift 70-80% of their work to open-source models to control runaway AI inference costs.
Chapters
I spent on 3.5 billion tokens this month myself. What do you think my spend was on Anthropic? 6 grand. 5.6. But in my mind, like, and Neil and I have talked about this on the phone, it's like we're gonna need to buy more infrastructure for our companies just because it's untenable. If I'm spending this much, I can— once people get to like even half the level that like I'm spending right now, it's— you can't afford it. So you need at least 80%, 70-80% is open source. So we— I don't have the numbers in front of me, but I know internally we've been looking at our cost for AI for our engineers and our product team. The cost has been exploding. It is becoming a massive cost center.