Robert Ta discusses how enterprises are hemorrhaging money on generative AI, with one company accidentally spending $500 million in a month. He questions Anthropic's $47 billion revenue claim, suspects uncontrolled spending inflates it, but says the market is still huge. Claude-code is hitting 1.5 million downloads daily as designers and PMs start shipping code.
Mystery company accidentally blew $500 million on cloud AI in a single month, failed to put usage limit on licensees for employees. A new report claims AI is getting too expensive for big companies. Also, for people that are not deep into the lore as we are, there are some rumors this might be Amazon, but this is what we mean is, hey, Anthropic's got this run rate of $47 billion annual recurring revenue, and they're about to IPO. But bridging into that. Are they actually making that much money or is it really a poor planning exercise on the part of the enterprises? As I say that, enterprises never plan well after having worked for some enterprises. So actually it's that, I think like just based on my experience, probably really high on tech executives don't know what to do with this. Like they don't know how to budget for it. They don't know how to optimize things. We're seeing things like token maxing and leaderboards, which obviously lead to blow up half billion dollar month spend on your tokens. So clearly the industry's got to figure out some costs and controls and attribution and stuff like that so that this doesn't happen. So do I think that their run rate is largely magnified by things like this, that $47 billion? And then further, do I think it will keep growing? I think probably yes, but I also think it'll keep growing. That's my thought, is I, I think there's just so much surface area of market that they haven't captured yet. And that's why they're activating these enterprise consultant ventures, uh, to deploy their technologies, the Frontier Labs, into enterprises. I think it's probably a yes and a yes. Like, it's ballooned a little bit, but I still think the TAM is so big that they still have more to go even if the balloon pops a little bit.
And I guess the wisdom has been that they're supply-constrained, you know, that they have way too much demand for the supply anyway, so that they're just going to be able to sell as much supply as they can sell for the foreseeable future anyway.
I listened to, uh, different podcasts and earlier this week it was reported that Claude code downloads had reached over 1.5 million a day. And, you know, some of that is like, you know, personal agents setting up their own Claude code, you know, that kind of thing. Uh, maybe not like real humans. If we're at 1.5 million a day, then if we have 8 billion people in the world, like at some point we're going to saturate, but, uh, I, that is an interesting number. That's increased from the few hundred thousand that was earlier this year, I believe.
Hmm.
Yeah.
It's incredible.
Yeah. So demand is definitely still there is what we're saying, even with this token budget optimization problem phenomena.
Yeah. Think about all the people who didn't write software before that, that now write software, you know, I've heard, you know, one informed estimate said there might be 10x the number of developers today. You know, or on the way to 10x as they were before, because you have people that like used to write code, but don't anymore. Now it's easier so that, so that, well, they have few people that always wanted to write code, but you know, like these stupid tools got in the way. You don't have to worry about that. You have designers, product managers, people on the periphery of, of, of using code all getting pulled in. Everybody's empowered. So yeah, no doubt the number of, of Claude, uh, code downloads is going up. Uh, they're creating new market for it. It's cool.
