Stanford CS 153: Infra @ Scale - Vercel CEO Guillermo Rauch

5 months agoMarch 10, 2025
37:16
10,564 views
366 likes
C

CS153 Infra at Scale

Channel

Interviewed Person

Guillermo Rauch

Description

Vercel CEO Guillermo Rauch talks about scaling and adapting on demand

Transcript

Group by:

early on you decided that you were going to build an infro layer on top of other info providers like AWS and a lot of you know a lot of folks in the in the infer Community were like well what's the point of that right why why do you need versel when you've got AWS so could you go take us back in history a second and and talk us through how you got to your current stack and where you guys sit in it yeah for sure so my my first experience with the cloud was actually a pretty tedious one so I wanted to just deploy a website and I wanted to by the way like the thing about versel that's

interesting is the ambition was not to just help people build websites and web applications there were a million ways of doing that I wanted to help people build the kinds of websites and applications that the Giants of the web were building so a little bit was like that Meme of like copying the homework like I actually wanted to like help you build like Google and Amazon and meta but I wanted to make it as easy as it is to build with freaking like square space or Wicks and that was novel it didn't

exist in fact a lot of investors told me it was impossible but there were new innovations that were coming into the cloud were starting to make it possible so for example our first version of versell used kubernetes as the place where the the the workloads actually run and then we realized you couldn't actually create a system that had the ambition of resell with that system it was too heavy it was too rigid

it was too static it scaled up and down quite slowly and so we had to learn how to design infrastructure that met that ambition of like hey it should be one click or one command to deploy so going back to your question or prompt about how is it possible well it turns out that the hyperscalers have created hundreds of services that customers need to choose from and they actually come to the market with a point of view of you start with the infrastructure

right log AWS console blah blah blah configure infra and then you put an application in it versel actually took the opposite they we inverted that process we said you start with the application here's a framework nextjs and there's others that give you guard rails that give it get give shape to that application and then when you deploy it we automated creation the infrastructure today we call that framework defined infrastructure it was actually inspired by softwar defined Network

and it was a similar Revolution but in the sort of like networking space we virtualized the cloud so it actually makes perfect sense that we build on dou AWS because Enterprises trust AWS they're leaving behind the data center but they cannot figure out how to get to the Cloud and that was not I think I kind of stumbled upon that and like it was a lot of work and you know it was uncertain whether Enterprises would want this but now Enterprise is actually I would argue even wanted more than developers but I started with if I'm a

developer and I have a new idea the infrastructure has to be invisible and what matters the most is the idea and the app code um there's a you know Meme now that uh that versel has started has has has accelerated right which is that um if you need anything in in the application layer you essentially just wait around long enough and versel was for you um why is that what what how did you go

from going okay we've got an nextjs framework here that's largely a web full stack framework then now suddenly you're shipping multi-agent Frameworks for you know we had Ben man from anthropic um an hour ago here you've got all the anthropic models that you're serving you've also got all the latest reasoning stuff what why uh is it just a natural progression of computing systems for you to go from nextjs they're also supporting uh everything at the uh uh AI layer yeah for us always about what do

75 segments (grouped from 351 original)6429 words~32 min readGrouped by 30s intervals

Video Details

Duration
37:16
Published
March 10, 2025
Channel
CS153 Infra at Scale
Language
ENGLISH
Views
10,564
Likes
366