Blackstone says Wall Street is complacent about AI disruption

ft.com

8 points·master_crab··4 comments

Building an AI-powered SaaS right now and the cost economics are wild. 18 months ago, running GPT-4 on every user interaction would've bankrupted most startups. Today it's $10-20/user/month and dropping. this is the disruption curve Blackstone is talking about. Once AI becomes "cheap enough," entire business models flip. Industries with high-margin knowledge work (finance, consulting, law) are sitting ducks. Wall Street sees the tech, but underestimates how fast costs drop + capabilities improve. We're in the iPhone moment of AI.

Are those $10-20/user/month costs sustainable? As I understand it, a lot of the price tags we're seeing now reflect steep discounts underwritten by massive debt and VC money.

The other thing I wonder about with respect to cost and availability, at least in the U.S., is the fact that our electrical grid is in shambles and there's a ton of deferred maintenance. Even if AI can deliver on its promises, do we even have the infrastructure in place to power it?

These are just two issues; I'm sure there are more, and I imagine Wall Street is taking some of this into account in their analyses.

Depends on which part we’re betting on survive the bubble pop. There’s a lot of capability being dismissed in low power inference while the large players push they need more money for large inference. After the pop people will start realizing how much benefit we can get without needing to create a Dyson sphere. More power is needed for continued major progress, but little power is needed for what’s going to stabilize as the normal use 95% of people need. Businesses will be users of the pricey inferences.

Only time will tell how things really play out. But these are my predictions as of today.

https://archive.is/W7Rnm

Reply on news.ycombinator.com