Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal CEO
October 22, 2026
AI Summary
5 min readWhy AI Compute Comes in Rectangles
Erik Bernhardsson, CEO of Modal, started his company because building AI infrastructure "fucking sucks." After spending seven years at Spotify building the first version of its music recommendation system—heavy on matrix factorization and compute—he wanted a tool that didn't exist: a way to run code in the cloud that was actually easy. Modal launched in 2021, and by November 2022 it was doing $100 a month in revenue. Then Stable Diffusion came out. The next month: $2,000. The month after that: $30,000. "I finally felt this massive anxiety thing off my chest," Bernhardsson says. "Maybe this is actually like we're onto something." Today Modal runs inference and sandboxes for companies like Suno, Runway, and Cognition, and has grown roughly 50x in revenue and users since his last public appearance two years ago.
The Capacity Problem: Rectangles Don't Fit Real Demand
Modal's core insight is that the GPU market has a structural mismatch. Almost all capacity is sold as "three-year rectangles"—fixed blocks of GPUs that start three to six months in the future. That is not what demand looks like for most users, whose workloads scale up and down over 24-hour cycles, burst unpredictably, and shift between models as new ones launch.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:03) **Company Overview & Origin Story** - Erik Bernhardsson explains Modal as an AI infrastructure platform for companies like Descript, Suno, and Runway, covering inference, sandboxes, and training.
- 2 (04:32) **The Path to Modal: From Workflow Orchestration to Compute Execution** - Erik describes how he iterated through several ideas before landing on Modal, including a web scraping tool and a workflow orchestration project.
- 3 (07:04) **Founder-Led Marketing & Early Revenue Trajectory** - Erik describes how his compulsive tweeting and blogging drove early growth, and shares the dramatic revenue inflection point.
- 4 (10:26) **Why Modal Built Its Own Stack (Not Just a Kubernetes Wrapper)** - Erik explains the thesis that new technology eras require rethinking tools from scratch.
- 5 (13:21) **The GPU Capacity Landscape: From Spot to 25 Clouds** - A detailed look at how Modal sources compute capacity and how the market has shifted.
- 6 (15:31) **Rent vs. Buy: The Financial Engineering of GPU Capacity** - Erik weighs the trade-offs of owning GPU hardware versus renting it.
- 7 (18:03) **Is GPU Supply Truly Constrained? A Counterintuitive View** - Erik challenges the narrative of a permanent GPU shortage.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
Warm welcome to Basil Chatha, our newest cohost from the Forward Deployed side of the AI Engineering world! We’ll be ramping up a new show with him and couldn’t pass up an interview with Modal’s CEO Eric on the occasion of their monster Series D!
Two years ago, we spoke with Erik Bernhardsson about his vision for truly serverless AI infrastructure: developers should be able to run code without provisioning clusters, predicting load, or paying for idle resources. CTO Akshat then joined us to talk about Modal’s infra primitives and stack to offer those primitives and agent experience.
A lot has changed in the past 5 months. Since their $5B series C, Modal now reports it crossed $1 billion in annualized run rate, raised a $750 million Series D at a $16B valuation, and has spun up nearly three billion sandboxes. In this conversation, Erik joins Vibhu to explain the infrastructure and economics behind that growth—and why demand for compute rarely looks like the contracts used to buy it.
Watch the full conversation on YouTube →
GPU demand isn’t rectangular
A typical capacity contract gives a customer a fixed number of GPUs for a fixed period. Plot capacity against time and you get a rectangle. Real workloads behave differently: they grow, spike, and fall throughout the day. Buy enough for the peak and you pay for idle hardware; buy less and you risk missing demand.
Erik describes Modal’s bet as a kind of duration transformation: pool demand across thousands of customers, secure capacity underneath, and let each customer use what they need when they need it. The smoother the aggregate demand, the more useful that pool can be. It’s a useful companion to our conversation with SF Compute’s Evan Conrad: Evan focuses on making capacity tradeable; Erik focuses on making it usable in smaller, elastic slices.
There is a financial risk on the other side. If a provider locks in capacity at today’s prices and GPU prices later fall, it can be left holding expensive supply. Erik says he doesn’t expect an imminent collapse, but thinks the possibility matters when building the business. Jump to the GPU price di