Modal

Serverless cloud for AI and data workloads that lets engineers run Python at scale on pooled CPUs and GPUs with sub-second container starts and per-second billing. Founded in 2021 by Erik Bernhardsson — former CTO of Better.com and creator of Spotify's recommendation system and the Annoy library — and Akshat Bubna, Modal exposes compute as a Python decorator: wrap a function, deploy it, and Modal handles containerization, scheduling, autoscaling, and GPU provisioning. The company raised an $87M Series B led by Lux Capital in September 2025 at a $1.1B post-money, and as of February 2026 was reportedly in talks for a new round at a $2.5B valuation.

Modal has become the default compute substrate for AI-native teams that don't want to manage Kubernetes or GPU reservations — a fast-growing alternative to AWS for the next generation of inference, fine-tuning, and agent workloads.

modal.com AI Compute and Infrastructure

United Statesprivate

At a glance

Total raised

$111M

last round series b

Headcount

51-200

approximate · public range

Headquarters

New York

United States

Founded

2021

5 years active

Status

Private

AI Compute and Infrastructure

Latest valuation

$1.1B

Series B · 2025

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

94 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$111M

Rounds tracked

Latest stage

Series B

Latest valuation

$1.1B

Date	Stage	Amount	Valuation	Lead investors
Sep 29, 2025	Series B	$87M	$1.1B	Lux Capital

Founder backgrounds

Akshat Bubna

Co-founder

Akshat Bubna studied mathematics and computer science at MIT (2014–2017) and was a competitive programmer who earned a Gold medal at the International Olympiad in Informatics (IOI) 2014 — the first Indian student to do so — as well as a Bronze at IOI 2013. Before co-founding Modal, he was an early employee and Staff Software Engineer at Scale AI, where he designed core engineering and operations systems and led the Natural Language team. That first-hand experience with the friction of running AI workloads at scale — watching teams struggle with Kubernetes and infrastructure plumbing — directly informed Modal's mission to build a serverless cloud that eliminates those same bottlenecks.

Erik Bernhardsson

Co-founder & CEO

An early engineer at Spotify (2008–2015), where he built the company's first data team and later ran music recommendations — contributing to the algorithms behind Discover Weekly. During that time he created **Annoy**, a widely used open-source library for approximate nearest neighbor search. From 2015 to 2021 he was CTO at **Better.com**, scaling the engineering organization through a period of rapid growth. He studied at KTH Royal Institute of Technology in Stockholm. His prior work building high-throughput recommendation infrastructure at Spotify directly informs Modal's focus on serverless, latency-sensitive GPU and CPU compute for AI/ML workloads.

AI bull / bear

AI bull / bear analysis hasn’t been generated for this company yet. Cases refresh quarterly using the latest classified articles and funding history.

In the news

The New Stack5/19/2026
Anthropic debuts MCP tunnels and self-hosted sandboxes to lock down AI agent infrastructure
Anthropic launches public beta self-hosted sandboxes and research-preview MCP tunnels for Claude Managed Agents, with support for infrastructure from Cloudflare, Modal, and Vercel.

Watch

Leadership

Erik Bernhardsson

Co-founder & CEO

in 𝕏

Akshat Bubna

Co-founder

—

Products

Modal Functions
Modal Functions is the core developer experience: decorate a Python function with @app.function, specify hardware (CPU, A100, H100, B200) and dependencies in code, and Modal handles containerization, deployment, autoscaling from zero, and per-second billing. Containers cold-start in under a second using a custom file system and snapshotting, making serverless GPU workloads viable for latency-sensitive inference. Functions can fan out to thousands of parallel containers for embarrassingly parallel jobs like batch inference, web scraping, or fine-tuning sweeps.
Modal Sandboxes
Sandboxes are short-lived, isolated Linux environments designed for running untrusted or AI-generated code at scale. Each sandbox boots in milliseconds, supports arbitrary languages and tools, and exposes a file system, shell, and network namespace through the Modal SDK. Sandboxes are heavily used by AI agent platforms and coding agents to execute model-generated code safely, enabling tools like code interpreters, autonomous agents, and online evaluation harnesses without operators having to run their own VM fleet.
Modal Inference
Modal Inference is the productized layer for serving open-source and custom models behind low-latency HTTPS endpoints. Engineers deploy models with a few lines of Python, get autoscaling GPU-backed serving with sticky sessions, batching, and KV-cache optimizations out of the box, and pay only for the compute they actually consume. It targets AI startups that need OpenAI-style inference economics for Llama, Mistral, Stable Diffusion, Whisper, and bespoke fine-tuned models without standing up a serving stack themselves.

Open roles

No live openings right now, or the job board API is temporarily unreachable. The company's careers page has the latest.

View all careers ↗