July 16, 2026
Pods, shots, and scheduler gossip
Scaling to 1M concurrent sandboxes in seconds
They spun up 1 million tiny workspaces fast—and the comments instantly dragged Kubernetes
TLDR: Modal says it rebuilt its platform so it can start 1 million isolated workspaces in less than a minute, a big deal for AI services that need huge bursts of activity. Commenters were impressed but instantly turned it into a debate over whether Kubernetes is the wrong tool and whether Modal’s approach holds up under pressure.
Modal says it can now launch 1 million isolated workspaces in under a minute, a wild brag in the race to power the new wave of AI tools. But the real popcorn moment is in the comments, where infrastructure nerds reacted like this was both a breakthrough and a subtweet aimed straight at Kubernetes, the popular system many companies use to manage apps. One reader called the new scheduler the most impactful system they’ve worked on, basically rolling out the red carpet for the project. Others read the post as a giant lesson in "use the right tool for the job"—translation: stop forcing one-size-fits-all software to do everything.
And yes, there was immediate pushback. One commenter casually strolled in with a very Hacker News-style flex: they scale super-fast virtual machines with a simple Go service and a SQLite database, as if spinning up huge fleets is just something they do before coffee. Another thread got deliciously skeptical, grilling Modal on whether this optimistic, everyone-decides-at-once approach falls apart when machines start filling up. In plain English: does this thing still work when the parking lot is almost full?
The mood was equal parts awe, interrogation, and nerd-sniping. There weren’t many outright dunks, but the vibe was unmistakable: impressive demo, now show the receipts. The funniest subtext? For a post about speed, the comments turned into a slow-burn custody battle over who really understands large-scale systems—and whether Kubernetes just got publicly roasted.
Key Points
- •Modal says it rebuilt its sandbox platform to support millions of concurrent sandboxes and tens of thousands of sandbox creations per second.
- •The company says it demonstrated the new platform by creating 1 million sandboxes in under a minute and running them concurrently.
- •The article argues that large-scale container platforms hit limits because many operations scale with node count, container count, or both.
- •Using Kubernetes as an example, the article cites serialized scheduling, O(n x p) worst-case scheduling, and etcd write pressure as scaling challenges.
- •Modal says its prior architecture also suffered from strong-consistency and global-coordination bottlenecks, including O(sandboxes) writes to Postgres and workflow/RPC load under high churn.