Cedric Clyburn
On Kubernetes Bytes
“They estimated, I think, six months ago, that your $200 subscription, like the Claude Max, would give you, as a personal user, about $20,000 of API tokens. … More recently, it's about $14,000. So you're kind of starting to see the amount of tokens you get for a specific plan go down. … Yes, these are definitely subsidized, but I think the subsidization is going down.”
Kubernetes Bytes · Hands-On Engineering Podcasts · November 2026
Context
Asked whether AI companies subsidize users, Clyburn cites a third-party industry estimate, from memory, of what a flat-rate coding subscription is worth at API prices, and says the subsidy is shrinking.
Transcript
…of weights and KV cash. It's going to be a lot of things. But the subsidization of inference is a really interesting conversation, both for developers, but also from an economic standpoint. There's a really good source that I read up on called Semi-Analysis. Have you heard of it before?
Nope, not me.
It's a very nerdy type of publication where they talk everything about inference and everything about model cost and tiers. And they estimated, I think, six months ago, that your $200 subscription, like the Claude Max, would give you, as a personal user, about $20,000 of API tokens. Yeah. So you could say they're subsidizing. Yeah, they definitely are. More recently, it's about to be November 2026. It's about $14,000. So you're kind of starting to see the amount of tokens you get for a specific plan go down. But remember, this is the API cost that they're charging external customers, right? Like anybody off the street. You have companies like Cursor who are reselling…
Yeah. You know, carrot is string. Yeah. The carrot is a stick. Right now, yeah, eventually we'll get this thing at some point in time. Once they have, I don't know, Anthropic goes public and they have their first quarterly earnings report and the stock tanks are like, okay, we need to bump up the…
Yeah, can't wait to see that. Well, I don't hope to see that, but I can't wait to see that too. True. So it's an interesting time.
Yeah, for sure. Okay. So now, let's say I've gone through this economic analysis and figured out that I want to run my own models. What are some of the SLOs that I need to think about before going and deploying something on Kubernetes on day one, right? Like, how do I plan for this? Can you talk…
More from Kubernetes Bytes
