Modal Labs

Company Snapshot

The $300M+ annualized-revenue figure is company-reported, not independently audited in the public source. The Series C valuation and funding are company-reported. citeturn1search1

1. Company Overview

Modal's own company page says its goal is to make it easier to iterate and ship applications for data, AI and machine learning, while its documentation describes serverless execution with per-second usage pricing. citeturn0search1turn0search2

2. Founders & Creators

Akshat Bubna

Co-FounderCTO

Bubna studied mathematics and computer science at MIT and was an early engineer at Scale AI. Secondary biographical research also records his International Olympiad in Informatics gold medal. The current Modal company page identifies him as a founder and the company's CTO. citeturn0search1turn2search1

Important classification: this report does not label either founder the sole “inventor” of Modal's infrastructure. Modal is a team-built system; specific components should be attributed to the engineers who developed them unless a source explicitly assigns individual authorship.

3. Founder Story & Origin

Modal began before the current generative-AI boom. Bernhardsson's experience operating data and ML systems at Spotify and a large engineering organization at Better exposed a recurring problem: cloud infrastructure was powerful but often required too much operational glue for data and ML teams.

Modal's official history says the founders wanted to make it easier to iterate and ship applications for data, AI and machine learning. Instead of building a thin wrapper around existing services, they went deep into the infrastructure stack. citeturn0search1

Secondary research describes an early period in which the team spent substantial time building foundational infrastructure before broad customer traction. Stable Diffusion and the explosion of generative AI then created a much larger demand for elastic GPU inference, helping accelerate product-market fit. This historical interpretation is secondary research and should not be treated as an audited company account. citeturn2search1

4. What Is Modal?

Beginner explanation

Think of Modal as a programmable cloud where you give the platform code and say what CPU, memory or GPU the job needs. Modal starts the compute, runs the code, scales it when demand changes and bills for usage.

Developer explanation

Modal turns ordinary Python functions into cloud functions and lets developers define container environments and hardware configuration in code rather than assembling a large YAML/Kubernetes deployment stack. Its current documentation also exposes JavaScript and Go SDK paths for several resources. citeturn0search2turn0search0

AI/ML engineer explanation

Modal provides elastic execution primitives for inference, distributed/batch computation, fine-tuning, GPU-backed notebooks and isolated execution environments. GPU selection currently includes NVIDIA T4, L4, A10, L40S, A100, H100, H200, B200 and B300 families, with availability and pricing subject to change. citeturn0search3

5. Core Technology & Architecture

Modal's 2026 serverless-GPU engineering post describes the platform as an AI-oriented cloud stack built to reduce startup latency and make GPU capacity more elastic. citeturn1search10

6. Products & Services

7. Modal Sandboxes

A Modal Sandbox is an isolated execution environment that can run arbitrary code. Modal specifically positions Sandboxes for code generated by language models, untrusted code, coding agents and reinforcement-learning environments. citeturn0search7

Current resource controls allow CPU and memory requests plus hard limits, and GPU Sandboxes are available subject to GPU scheduling/preemption constraints. citeturn0search0

Modal's next-generation Sandbox backend is currently documented as beta and is recommended for workloads above 20 sandbox creations per second or more than 10,000 concurrent sandboxes. citeturn0search13

8. GPU Infrastructure

Modal's current GPU documentation lists T4, L4, A10, L40S, A100, H100, H200, B200 and B300 options, with support for multiple GPUs per container on several of the listed families. citeturn0search3

Modal also runs containers across multiple clouds and lets users specify regions for latency-sensitive workloads. citeturn0search14

Functions are preemptible by default; long-running workloads should tolerate interruption. Non-preemptible functions can be requested for CPU/memory, but the current documentation says that option is not supported for GPU Functions. citeturn0search15

9. Pricing & Business Model

Modal's core model is usage-based: the platform hosts the infrastructure and charges by usage rather than requiring users to manage servers. The public product pages advertise $30/month of free compute for new users. citeturn1search8

10. Funding History

11. Customer Case Studies

12. AI Use Cases

13. Technical Innovations

14. Modal vs Competitors

These are positioning comparisons, not claims that each competitor was directly evaluated by every Modal customer.

15. Modal vs Replicate — When Should a Developer Choose Modal?

16. Security & Compliance

Modal publicly advertises team controls, isolation, SOC 2, HIPAA and data-residency controls. Its Sandbox documentation also emphasizes isolated execution and resource limits for potentially untrusted code. citeturn1search8turn0search7turn0search0

For GPU Sandboxes and long-running GPU functions, preemption and availability constraints remain operational considerations. Organizations with strict regulatory or data-location requirements should verify the exact current region, retention and contractual controls applicable to their workload rather than relying on general product claims. citeturn0search15turn0search14

17. Developer Experience

• Python-first workflow and code-defined infrastructure.
• No-YAML positioning for core application deployment. citeturn0search2
• GPU selection through simple code configuration.
• Serverless execution with automatic scaling.
• CLI, SDKs and examples for common AI workloads.
• Notebooks, Batch and Sandboxes extend the same compute model beyond inference.

The core product philosophy is to let developers write the application and treat infrastructure as a programmable substrate rather than a separate operations project.

18. Company Timeline

19. Financial & Growth Information

20. Challenges & Limitations

• GPU availability: AI infrastructure depends on scarce accelerator capacity and regional supply.
• Preemption: default preemptibility means long-running jobs must be designed for interruption tolerance. citeturn0search15
• Cold starts: AI workloads often require large model weights; reducing startup latency is a continuing infrastructure problem.
• Cost: elastic serverless economics can reduce idle capacity but sustained GPU workloads still produce significant compute bills.
• Multi-node complexity: large-model inference/training requires high-bandwidth networking and careful distributed execution.
• Cloud competition: Modal competes with hyperscalers, GPU clouds and specialized inference platforms.
• Platform dependency: customers outsourcing infrastructure gain speed but accept dependency on Modal's availability, pricing and product roadmap.

Controversies: No major legal or public controversy was identified in the authoritative Modal sources reviewed for this profile. That is not evidence that no criticism exists anywhere; it means no material controversy was substantiated in the source set used here.

21. SWOT Analysis

22. Future Outlook

Confirmed direction

Modal's 2026 communications emphasize low-latency elastic inference, dynamic agent runtimes, reinforcement learning, batch workloads and safe isolated execution. The Series C announcement explicitly frames the company as a broader AI cloud rather than a single-purpose GPU provider. citeturn1search1

Evidence-based analysis

The most important strategic opportunity is to become the infrastructure layer for companies that increasingly own their own models and post-training loops. Modal's customer examples already span model serving, RL, robotics, biology, media and agents, suggesting the platform is deliberately broadening beyond conventional inference. citeturn1search1turn1search0

Prediction, not confirmed fact: if agentic applications continue to grow, Sandboxes, batch execution, RL environments and low-latency model serving could converge into a unified “agent compute” platform. The main risk is that hyperscalers and specialized AI clouds may replicate these primitives quickly.

23. 50 Important Facts

1. Modal's legal name is Modal Labs, Inc.
2. Modal was founded in 2021.
3. Erik Bernhardsson is a co-founder.
4. Akshat Bubna is a co-founder.
5. Bernhardsson is CEO.
6. Bubna is CTO.
7. Modal is headquartered in New York City.
8. The company focuses on AI, ML and data infrastructure.
9. Modal uses serverless execution.
10. Pricing is usage-based.
11. Modal supports CPU workloads.
12. Modal supports GPU workloads.
13. Current GPU docs list H100.
14. Current GPU docs list H200.
15. Current GPU docs list B200.
16. Current GPU docs list B300.
17. Modal supports multiple GPUs per container on supported hardware.
18. Modal supports training.
19. Modal supports fine-tuning.
20. Modal supports inference.
21. Modal supports batch processing.
22. Modal supports notebooks.
23. Modal supports Sandboxes.
24. Sandboxes can execute AI-generated code.
25. Sandboxes have CPU/memory limits.
26. GPU Sandboxes can be preempted.
27. Modal supports region selection.
28. Modal runs containers across multiple clouds.
29. Modal has its own container runtime.
30. Modal has its own scheduler.
31. Modal has its own filesystem.
32. Modal has its own image builder.
33. Modal emphasizes code-defined infrastructure.
34. Modal's docs explicitly market a no-YAML experience.
35. Modal Batch launched in 2025.
36. Modal Batch was designed for large fan-out workloads.
37. Harvey reported a 10× speedup using Modal Batch.
38. Suno reported four months saved on launch timing.
39. Suno uses thousands of GPUs on Modal.
40. Physical Intelligence reported about 10–15 ms network overhead.
41. Runway reported moving from proof of concept to production in under 30 days.
42. Modal raised $16M Series A in 2023.
43. Modal raised $87M Series B in 2025.
44. Series B valuation was $1.1B post-money.
45. Modal raised $355M Series C in 2026.
46. Series C valuation was $4.65B post-money.
47. Modal reported more than $300M annualized revenue in May 2026.
48. Modal's customer examples span AI, biotech, robotics and media.
49. Modal's current strategy emphasizes AI-native infrastructure rather than a single model API.

24. Lessons for Entrepreneurs & CTOs

1. Build around a painful workflow, not a fashionable model. Modal started with developer infrastructure pain before generative AI became mainstream.
2. Own the hard layer when abstraction quality is the product. Modal built core infrastructure rather than merely wrapping a cloud API.
3. Serverless economics are especially useful for spiky AI workloads. Suno's holiday demand is a clear example. citeturn1search3
4. Developer experience can be a moat. Code-defined infrastructure reduces the distance between prototype and production.
5. AI infrastructure should cover the lifecycle. Inference alone is increasingly insufficient; training, batch, notebooks and agents all interact.
6. Security becomes a product feature when agents execute code. Sandboxes are infrastructure for safe execution, not just developer convenience.
7. Latency is application-dependent. Robotics and real-time video need different infrastructure from offline batch processing.
8. Infrastructure companies need broad primitives but clear interfaces. Modal's strategy is to expose a small developer-facing surface while operating complex infrastructure underneath.

25. Final Verdict

Sources & Verification

1. Modal — Company
2. Modal — Documentation
3. Modal — GPU acceleration
4. Modal — Sandboxes
5. Modal — Sandbox resources and pricing
6. Modal — Series B
7. Modal — Series C
8. Modal — General availability and Series A
9. Modal — Suno case study
10. Modal — Physical Intelligence case study
11. Modal — Runway case study
12. Modal — Batch launch and customer examples
13. Modal — Customers
14. Modal — Serverless GPUs technical post
15. Modal — Notebooks
16. AI Files — Erik Bernhardsson profile
17. TechCrunch — February 2026 funding discussions
18. Contrary Research — Modal founding history