Building AI-Native Infrastructure for Developers | Erik Berhnardsson, CEO of Modal
The Peel with Turner Novak
Turner Novak
4.6 • 11 Ratings
🗓️ 16 October 2025
⏱️ 101 minutes
🧾️ Download transcript
Summary
Erik Bernhardsson is the Co-founder and CEO of Modal, building high-performance AI infrastructure.
We talk about building what is essentially a new cloud provider, created from the ground up, optimized for AI. We also talk about what actually happened with the great GPU shortage, how Modal fixes the inference problem in AI, and why he thinks AI will lead to 10x more developers.
He also shares lessons on culture from joining Spotify as the 40th employee, treating hiring like a prediction problem, what most people get wrong working with early customers, why more people should start companies in their 30’s and 40’s, and reflections on fundraising in a hot market.
Thank you to Tim Chen at Essence Venture Capital and Erik’s Co-founder Akshat for their help brainstorming topics for this conversation.
Thank you to Meow and Hanover Park for supporting this episode.
Meow: Get free bookkeeping for your startup at https://www.meow.com
Hanover Park: Modern, AI-native fund admin at https://www.hanoverpark.com/Turner
Timestamps:
(4:26) Modal: AI-native infrastructure
(9:02) Why its so hard to get GPU’s
(15:00) Hitting PMF with AI generated media
(20:37) Competing in IOI competitions
(23:09) 40th employee at Spotify
(27:17) Lessons from Spotify
(31:17) Starting Better[dot]com
(34:05) Treating hiring like a prediction problem
(36:12) Erik’s favorite interview question
(39:07) Sales + common design partner mistakes
(42:02) Startups should solve hard problems
(44:05) Evolution of Modal’s product over time
(50:15) Rise in importance of inference in AI
(52:07) AI development post-GPU scarcity
(58:51) Building a brand in dev tools
(1:04:31) Fundraising from Seed to Series B
(1:07:42) More 30+ year old’s should start companies
(1:10:00) Reducing developer tax, increasing productivity
(1:20:37) Why Erik’s bullish and bearish on AI
(1:26:17) Bubbles, downsides to inappropriate valuations
(1:34:58) High CO2 levels make you dumb
(1:37:38) Difference between US and European startups
Referenced
Modal: https://modal.com
Careers at Modal: https://jobs.ashbyhq.com/modal
Suno: https://suno.com
Planet Scale: https://planetscale.com
How to hire smarter than the market: https://erikbern.com/2020/01/13/how-to-hire-smarter-than-the-market-a-toy-model
Interviewing is a noisy prediction problem: https://erikbern.com/2018/05/02/interviewing-is-a-noisy-prediction-problem
Cloud in 2030: https://erikbern.com/2021/11/30/storm-in-the-stratosphere-how-the-cloud-will-be-reshuffled
Follow Erik
Twitter: https://x.com/bernhardsson
LinkedIn: https://www.linkedin.com/in/erikbern
Follow Turner
Twitter: https://twitter.com/TurnerNovak
LinkedIn: https://www.linkedin.com/in/turnernovak
Subscribe to my newsletter to get every episode + the transcript in your inbox every week: https://www.thespl.it/
Transcript
Click on a timestamp to play from that location
| 0:00.0 | We had to throw out all the existing infrastructure, |
| 0:02.4 | kind of rethink the entire developer experience, |
| 0:05.2 | and rethink a lot of managing containers, |
| 0:08.4 | managing large models and working with GPUs. |
| 0:11.2 | So the core idea is like, we don't build the applications, |
| 0:13.8 | we build like the infrastructure below, |
| 0:15.4 | and we're building kind of a new infrastructure stack |
| 0:17.2 | that's more well suited for AI. |
| 0:19.0 | Welcome to the PEO! |
| 0:20.4 | I'm your host, Turner Novak, founder of Banana Capital. Today's guest is Eric Bernhardtson, co-founder and CEO of Model. We talk about building what is essentially a new cloud provider created from the ground up, optimized for AI. 90% of the workers is building infrastructure, not application. What actually happened with the great GPU shortage? |
| 0:54.9 | We talked to a lot of these startups. They're like, sitting on 1,000 GPUs. We can't leverage all of them. Like, we don't know what to do with them. Lessons joining Spotify is the 40th employee. If you just have smart people who are pretty autonomous and have a mission, you need a lot less structured than you think. People should just come in every morning and just ask themselves, what can I do for the business today? |
| 0:57.1 | Why you should treat hiring like a prediction problem? |
| 0:57.1 | Like being a little impatient. I like people who want to just get stuff done. What to prioritize |
| 1:01.8 | during hypergrowth? I mean, when you have five engineers and like you find product market fit, |
| 1:06.4 | you don't have a choice. Just like chase that shit. What most people get wrong working with early |
| 1:10.8 | customers. |
| 1:11.7 | You should take their pain points extremely seriously. |
| 1:14.4 | But when they suggest solutions, you should almost ignore that. |
| 1:18.0 | How Modal fixes the inference problem in AI. |
| 1:21.7 | Unlike training, where you're actually fine having a fatic amount of GPUs. |
| 1:25.6 | When you're doing inference, you have very volatile demand. Why you should start a company in your 30s and 40s? It's also hard to find a problem that's hard, but not impossible. And like knowing that. How AI is changing software and developer productivity. I think there's so much software that has to get built. I think we're going to need a lot more engineers over the next decades. And thoughts on raising capital in hot markets. The one clear misalignment I see in like |
| 1:49.6 | VC versus founders is like DCs will always advise you to take more money than you need. A quick |
... |
Please login to see the full transcript.
Disclaimer: The podcast and artwork embedded on this page are from Turner Novak, and are the property of its owner and not affiliated with or endorsed by Tapesearch.
Generated transcripts are the property of Turner Novak and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.
Copyright © Tapesearch 2026.