Modal CEO: Has Open-Source LLM Inference Peaked? | What Happens to the Cost of Compute
Modal CEO Erik Bernhardsson joins Harry Stebbings to discuss the inference market, GPU buying, margins and hiring. He argues open-source inference has not peaked as Modal reaches $1 billion in annual revenue.
The gist
-
Bernhardsson says selling only tokens is a race to the bottom, so Modal bundles inference with sandboxes, post-training and batch jobs.
-
Stebbings says he runs over 100 automations on Instinct, and both agree compute demand likely can't be overbuilt in five years.
-
Modal's sandbox product launched in 2023 and took off in summer 2025, and is now a $300 million run rate, about a third of revenue.
-
Bernhardsson rejects the claim that open-source inference has peaked, predicting both open and frontier model usage will grow 5x to 20x in 12 months.
-
He expects a 20% to 30% market pullback but stays bullish, and warns against companies that depend on constant new equity funding.
Harry Stebbings uses it
"I run Instinct for everything. I have over 100 automations on Instinct doing things that are just absolutely wild, from buying concert tickets to planning shows to helping with medication for my mother."
▶ Play at 21:1618 creators recommend it, including Kevin Rose, Shaan Puri, and Sheel Mohnot.
See all on the Instinct pageWhat they recommend · 1 pick
Instinct
AI tools
"I run Instinct for everything. I have over 100 automations on Instinct doing things that are just absolutely wild, from buying concert tickets to planning shows to helping with ..."
Also mentioned
1:52
Mac
· "I realized you can control computers. I had a Mac Plus, you know, 1 megabyte of RAM or whatever. And yeah,..."
8:07
7 Powers
· "Moat to me is like, you go read Seven Powers or one of those things. It's a good book, but I think people..."
10:57
Fireworks AI
· "What has Modal done then that Baseten and Fireworks AI haven't to build that?"
16:35
Nvidia GTC
· "I spent some time in March at GTC, Nvidia's big conference in San Jose. I would describe that as like an..."
17:35
Kubernetes
· "We built a lot of our infrastructure ourselves. We don't use Kubernetes and Docker and things like that. We..."
◎
Related episodes
Episodes covering similar topics
Inference Will Be $10TRN Market: What No One Sees | Software Investors are F***ed | Naveen Rao
ai infrastructure
Amazon Has Warehouses Of GPUs That Can’t Be Installed + Shocking Campaign Allegations
ai infrastructure
How Export Controls Helped Not Hurt China & Power is the Bottleneck to AI | Perplexity CEO
ai infrastructure
More from 20VC with Harry Stebbings
David Tisch: Why No Manager Wants to Stay Small & Why Competing with Mega Platforms is a Losing Game
Oct 31, 2026
Instinct Worth 30BN? | Microsoft Cuts 27,000 Jobs While Meta Taps Equity Markets for $95B CapEx
Oct 29, 2026
Satya Nadella on Open vs Closed Models: Who Wins | Why Companies Need to Own Their Own Intelligence
Oct 26, 2026