Erik Bernhardsson
@bernhardsson
Building everyone's favorite AI infrastructure platform @modal
4.1K Following 58.3K Followers
100% @modal 👑 arming the jevolution with GPUs. Y'all have been the best this week.
The most valuable knowledge is the one that requires long feedback loops to learn. Like months or years. It’s funny how this was always true but now AI makes this even more true.
I'm happy for people who sell their startups but personally I have zero interest because I would just go build Modal 2 anyway
I strongly believe the Federal Reserve should replace the inflation targeting with a direct GPU price target.
Keep jacking up rates until I get get B300s at $4/h again.
So much of our alpha is in the infra layer anyway. So we’d rather work with open source projects and contribute back. Thanks @sgl_project for building amazing stuff!
Want to meet some cool people? Want to hear about the future of AI infrastructure? Are you in SF on Oct 1?
👇
Runtime speaker lineup is live!
We're bringing together experts covering AI infrastructure, applications of AI in science and robotics, the future of software engineering, and more.
Show more
Hot take but I think every time a company picks a marketing slogan, they need to use it for at least five years
Honestly our design team is so core our identity at Modal. We have a high bar for the product itself, but also the website, the swag, the events, etc. It all reinforces that high bar, especially in a world of too much corny boomer slop.
Show more
last week I hit my 3 years at @modal where i've been leading design across product + brand and slowly building out our team. we've been investing more in brand as the company has grown and i'm excited to announce that @studiojud has joined us as our Director of Brand.
we worked together last fall as Studio Jud's first client, and i've spent the last year slowly convincing him to join 💚
...and we’re hiring more designers:
Show more
The biggest downside of the GPU scarcity is not that it's hard to get GPUs (you just have to pay more), but that almost all capacity is sold as a 3y+ "rectangle" starting 3-6 months into the future. That's not what demand looks like for most users!
Show more
Very cool work by one of our interns
And there's more cool FS perf coming up! Rendezvous hashing, RDMA, and much more.
I just wrapped up my internship @modal.
This summer I worked on container filesystems, speeding up agentic workloads by up to 4x and directory traversals by 12x. I wrote about this systems work behind it here:
Huge thanks to @amitpr, @jonobelotti_IO, the runtime team, and everyone at Modal for an amazing summer!
Show more
I just talked to @max_rimpel about building a new kind of cloud from scratch and why building AI applications forces us to rethink infrastructure.
"If you're not tackling hard problems, why are you creating a startup?"
@bernhardsson was told to use Kubernetes. Instead he built @modal's file system and container runtime from the ground up, tuned for the speed AI workloads demand. With Modal, containers boot in under a second, and workloads burst from zero to thousands of CPUs or GPUs in a few lines of code.
Watch @max_rimpel in conversation with Erik.
Chapters
00:00 — Learning to Code on a Mac Plus
04:38 — Building the Engine Behind Spotify’s Discover Weekly
09:17 — The Pandemic Project That Became Modal
10:14 — The 1 to 2 Year Rule for Picking Startup Problems
12:36 — Broke Until Stable Diffusion Hit
15:42 — Architecting a New Cloud for AI
17:41 — Agents Want the Same Things Developers Do
Show more
Very cool writeup by @mirrorphysics about how they are using @modal for drug discovery. We're seeing a ton of growth in computational bio.
Who knows, maybe one day I'll be taking drugs that were found on a Modal GPU.
Show more
found out we're serving 40% of Kimi K3 traffic on OpenRouter ...
the guy working on it is tired. anyone wanna help us work on inference?
👀
I'm doing some light renovation and the contractor keeps emailing about "Manhattan project" which feels a bit... confusing
Another week, another SotA! Hold @modal’s beer and we’ll come back with the best TPS on this one.
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog:
Show more
“Only in SF”
Then sharing something that definitely happens in 25 other cities on a daily basis.
Polymarket: Will Modal or Databricks IPO first?
I love having kids. They are on an island in Sweden with their grandparents for a week. I’m locking in. Calling them every day though.
it is qwensday my dudes
