Modal (platform)
Modal is a serverless cloud computing platform that lets developers run compute-intensive artificial intelligence, machine learning, and data-processing workloads on cloud GPUs by writing ordinary Python, with the platform handling containerization, scaling, and provisioning automatically.[1] Developers annotate standard Python functions with Modal decorators and the platform runs them in the cloud with no servers, Kubernetes, or Docker configuration to manage, eliminating the slow build-push-trigger-download loop that Modal founder Erik Bernhardsson describes as taking "a few minutes if you're lucky, or a few hours if you're unlucky."[1] Modal is developed by Modal Labs, Inc., headquartered in New York City with offices in San Francisco, Stockholm, and London.[9]
Founded in January 2021 by Erik Bernhardsson and Akshat Bubna, Modal has raised over $1.2 billion in venture capital funding and was valued at $15.75 billion after its October 2026 Series D.[9][21] The company reached a $1.1 billion valuation (unicorn status) with its September 2025 Series B round, then raised a $355 million Series C in May 2026 at a $4.65 billion post-money valuation and a $750 million Series D led by Accel in October 2026 at $15.75 billion post-money.[2][18][21] The platform serves thousands of customers including Meta, Scale, Substack, Ramp, and Lovable.[2] Its annualized revenue surpassed $300 million by May 2026, up roughly fivefold from about $60 million the previous September,[18] and in October 2026 Modal said it had crossed $1 billion in annualized run rate and had spun up close to 3 billion sandboxes.[21]
What is Modal used for?
Modal is used to run AI and ML workloads that need bursty access to GPUs without managing cloud infrastructure: serving model inference as autoscaling endpoints, fine-tuning and training models, running large-scale batch jobs, and spinning up sandboxed environments for untrusted code. Approximately 90% of workloads on the platform are related to AI and ML applications. Because billing is per-second and functions scale to zero, the platform targets the irregular, GPU-heavy usage patterns typical of data and ML teams rather than the steady traffic of traditional backend services.[1][6]
History and Founding
Modal was founded in January 2021 by Erik Bernhardsson, who serves as CEO.[1][11] Akshat Bubna joined as co-founder and CTO in August 2021. The company is legally incorporated as Modal Labs, Inc.[9]
Erik Bernhardsson
Bernhardsson is a Swedish software engineer who spent seven years at Spotify (2008 to 2015), where he built the core of the music recommendation system and built up and managed its machine learning team.[11][22] During his time at Spotify, he created two widely used open-source projects: Luigi, a Python workflow scheduler for building complex pipelines of batch jobs (over 17,000 stars on GitHub), and Annoy (Approximate Nearest Neighbors Oh Yeah), a C++/Python library for fast approximate nearest neighbor search in high-dimensional spaces.[16][17] Annoy's README says it is used at Spotify for music recommendations.[17]
After Spotify, Bernhardsson served as CTO of Better.com from February 2015 to January 2021,[22] where he scaled the engineering team from 1 to 300 people and built AI/ML systems for mortgage processing.[11] He began developing Modal during the COVID-19 pandemic, motivated by his repeated frustration with the gap between local development and cloud execution in data and ML workflows.[1]
Akshat Bubna
Bubna won a gold medal at the International Olympiad in Informatics (IOI) in 2014, becoming the first Indian student to achieve this distinction. Before joining Modal, he worked as an early engineer at Scale AI.[11] Bubna is Modal's CTO; the company says its team includes international olympiad medalists.[9][11]
Early Development
Bernhardsson's core insight was that data and ML teams have fundamentally different infrastructure needs from traditional backend engineering teams.[1][6] They require frequent infrastructure changes, bursty compute demands, and access to specialized hardware like GPUs and high-memory systems. Existing tools forced a slow cycle of building containers, pushing them to a registry, triggering jobs, and downloading logs, which destroyed the fast feedback loop that productive development requires.[1]
Rather than building on top of existing systems like Docker and Kubernetes, Bernhardsson decided to build Modal's infrastructure from scratch.[1][6] The first two years of development focused on foundational components: a custom container runtime, a proprietary filesystem, a scheduler, and an image builder, all written in Rust for performance.[6] The company operated in a closed beta during this period.[1]
Modal's first significant traction came when Stable Diffusion launched in August 2022, as developers needed a fast way to run GPU-intensive image generation workloads without managing infrastructure.[6] This validated the platform's serverless GPU model and attracted its initial wave of users.[6]
Funding and Valuation
Modal has raised over $1.2 billion in total venture capital funding across rounds from seed through Series D.[9][21] It reached unicorn status in September 2025, a $4.65 billion valuation in May 2026, and a $15.75 billion valuation in October 2026.[2][18][21]
| Round | Date | Amount | Lead Investor | Total Raised | Valuation |
|---|---|---|---|---|---|
| Seed | Early 2022 | $7 million | Amplify Partners | $7 million | Not disclosed |
| Series A | October 2023 | $16 million | Redpoint Ventures | $23 million | ~$154 million (post-money) |
| Series B | September 2025 | $87 million | Lux Capital | ~$111 million | $1.1 billion (post-money) |
| Series C | May 2026 | $355 million | General Catalyst, Redpoint | ~$466 million | $4.65 billion (post-money) |
| Series D | October 2026 | $750 million | Accel | ~$1.2 billion | $15.75 billion (post-money) |
Other investors in the earlier rounds include Lux Capital, Definition Capital, and Creandum.[3][9] The Series A coincided with Modal's general availability launch in October 2023.[3][4]
In February 2026, TechCrunch reported that Modal was in talks with General Catalyst to lead a new funding round at a valuation of approximately $2.5 billion, which would more than double the company's valuation from less than five months prior.[5] CEO Erik Bernhardsson characterized the discussions as general conversations rather than active fundraising.[5] At the time, Modal's annualized revenue run rate was reported to be approximately $50 million.[5]
How much is Modal worth in 2026?
As of October 2026, Modal was valued at $15.75 billion after a $750 million Series D led by Accel, with Kleiner Perkins joining and all existing investors participating; Modal said usage had tripled since its Series C.[21] Earlier, on May 21, 2026, Modal announced a $355 million Series C round at a $4.65 billion post-money valuation, roughly quadrupling the unicorn valuation it had set eight months earlier.[18][20] The round was led by General Catalyst and Redpoint, with Menlo Ventures, Bain Capital Ventures, and Accel joining as new investors and the company's existing major backers also participating.[18][19] According to Reuters reporting summarized by Tech Startups, the financing came together in two tranches: an initial group of investors backed Modal at a $2.5 billion valuation, and additional investor demand then pushed the second tranche to the $4.65 billion mark.[20] Modal said its annualized revenue had grown roughly fivefold since September 2025 to surpass $300 million, implying a starting point of about $60 million the previous autumn.[18] The company described the proceeds as funding for continued buildout of what it calls the production cloud for AI, with a focus on the low-latency inference, reinforcement learning infrastructure, and Sandbox capabilities used by AI agents.[18]
The company had approximately 14 employees at the time of its Series A in late 2023,[4] more than 120 by May 2026, and more than 200 by October 2026.[18][21]
Platform and Technology
Core Architecture
Modal's platform is built on custom infrastructure written primarily in Rust, designed from the ground up to minimize container startup times and maximize GPU utilization.[6][7] The architecture differs substantially from standard container orchestration platforms in several ways.
Custom Container Runtime: Modal uses Google's gVisor as its container runtime instead of standard Docker containers.[7] gVisor is a sandboxed container runtime that intercepts application system calls and acts as a guest kernel, providing stronger isolation than traditional containers that share the host system's kernel.[7] According to Amplify Partners, Modal was the first company to try running gVisor on machines with GPUs; at the time, GPU support in gVisor was highly experimental, and Modal has since contributed improvements to make gVisor with GPUs more production-ready.[6]
FUSE-Based Filesystem: Modal built a custom filesystem using FUSE (Filesystem in Userspace) with content-addressed storage.[7] Files are hashed and stored by their hash value, which eliminates duplication across container images.[7] When a container starts, it does not need to pull an entire image; instead, it lazily fetches only the files that are actually accessed.[7] Modal's data shows that applications typically access only a small fraction of their container image contents.[7] An in-memory index tracks file hashes, and an aggressive local SSD cache (with approximately 100 microsecond latency) stores frequently used files, compared to approximately 2 millisecond latency for network storage.[6][7] The filesystem was initially prototyped in Python, then rewritten in Rust for production performance.[6]
VolumeFS: A second custom filesystem called VolumeFS provides persistent storage for Modal functions across any region.[6] It uses eventual consistency rather than strict immediate consistency, which enables scalable distributed training and large dataset operations without sacrificing throughput.[6]
Automated Resource Solver: Modal implements an automated GPU resource allocation system using Google's OR-Tools Linear Optimization Package (GLOP).[6] This solver runs every few seconds, analyzing all active workloads and available GPU capacity across AWS, Google Cloud Platform, and Oracle Cloud Infrastructure.[6] It factors in regional constraints, pricing, and capacity to perform both comprehensive and incremental optimization passes, allowing the platform to elastically scale to hundreds of GPUs without users managing any cluster infrastructure.[6]
How fast are Modal's container cold starts?
Modal's stated goal is sub-second container startup.[7] Modal's pricing page says it spins containers up and down in 1-2 seconds, and Amplify Partners reported that memory snapshotting brings most cold starts below 3 seconds even with large dependencies; pulling large Docker images, by contrast, can take several minutes.[6][7][8] The platform achieves this through several techniques:
- Bypassing Docker entirely, pointing gVisor directly at a root filesystem with a JSON configuration rather than pulling container images[7]
- Content-addressed caching that avoids re-downloading files already present on the host[7]
- Lazy file loading that fetches only the small fraction of image files actually needed at runtime[7]
- CRIU (Checkpoint/Restore In Userspace) memory snapshotting, which captures the state of a running container and restores it on different machines, reducing initialization time for containers with heavy dependencies like PyTorch[6]
Python SDK and Developer Experience
Modal's primary interface is a Python SDK that uses decorators to define cloud infrastructure.[1][15] Developers annotate their Python functions with Modal decorators, and the platform handles containerization, environment setup, and remote execution automatically.[15] There are no YAML files, Dockerfiles, or Kubernetes manifests required.[15]
A basic Modal function looks like this:
import modal
app = modal.App("example")
@app.function()
def square(x):
return x ** 2
Environment configuration is expressed in Python rather than configuration files:
@app.function(
image=modal.Image.debian_slim().pip_install(["numpy", "pandas"]),
gpu="A100",
timeout=300,
)
def train_model(data):
# This runs on an A100 GPU in the cloud
...
Key SDK features include:
| Feature | Description |
|---|---|
@app.function() | Converts a Python function to a cloud-executed function with configurable resources |
image= | Defines the container environment using Python code (pip packages, apt packages, custom Docker images) |
gpu= | Specifies GPU type and count (e.g., "A100", "H100:2", "B200:8") |
schedule= | Sets up cron-like scheduled execution (e.g., modal.Period(hours=1)) |
.map() | Distributes work across multiple containers in parallel |
@app.cls() | Defines a class with lifecycle hooks and shared state across function calls |
@modal.fastapi_endpoint() | Deploys a function as an HTTPS endpoint (replaces the deprecated @modal.web_endpoint()) |
modal.Volume() | Attaches persistent storage to functions |
modal.Secret() | Securely manages and injects environment variables and credentials |
As of November 2026, Modal's SDKs for JavaScript/TypeScript and Go are in beta; they let projects in those languages call Modal Functions and use Sandboxes, while defining Functions remains a Python feature.[23]
Products and Features
Inference
Modal enables developers to deploy and scale LLM inference, audio models, image generation models, and other AI models as serverless endpoints. The platform provides instant autoscaling from zero to hundreds of GPUs and back down to zero, with per-second billing.[8] Users bring their own models and frameworks (such as vLLM, Hugging Face Transformers, or custom serving code), and Modal handles the underlying infrastructure.
Training and Fine-Tuning
The platform supports model training and fine-tuning on single-node or multi-node GPU clusters. Users can provision A100, H100, or H200 GPUs instantly without reservation or quota management. Modal's per-second billing means users pay only for the time their training job is actively running.[8]
Sandboxes
Modal Sandboxes provide isolated execution environments for running untrusted code from LLMs or users. Built on gVisor's security isolation, sandboxes allow developers to programmatically spin up ephemeral containers with controlled resource limits and filesystem access. Meta used Modal to spin up thousands of concurrent sandboxed environments for reinforcement learning when building its Code World Models (CWM).[2] Modal reported more than 1 billion sandboxes launched by May 2026 and close to 3 billion by October 2026, when it said Sandboxes had become a $300 million run-rate product on their own.[18][21] Cognition CEO Scott Wu has described the breadth of the workload: "Modal powers both our reinforcement learning infrastructure and production inference. Millions of sandboxes on one end, real-time serving on the other."[18]
Batch Processing
Modal supports large-scale batch processing workloads that can scale to thousands of containers on demand. The .map() and .starmap() operations distribute work across containers in parallel, with automatic retry logic and error handling. Common use cases include data preprocessing, web scraping at scale, and ETL pipelines.
Web Endpoints
Any Modal function can be deployed as a serverless HTTPS endpoint with a single decorator. These endpoints support standard HTTP methods and can serve as REST APIs, webhook handlers, or frontend-facing services. Modal handles TLS termination, load balancing, and autoscaling for these endpoints.
Scheduled Functions
Modal supports cron-like scheduling through the schedule= parameter in function decorators. Scheduled functions can run at fixed intervals or on custom cron expressions, enabling periodic data processing, model retraining, or monitoring tasks.
Notebooks
Modal previously offered real-time collaborative notebooks that ran on cloud GPUs. The product has since been shut down: Modal's documentation says users can no longer create, view, edit, or run notebooks in its editor, though existing notebooks can be downloaded as .ipynb files.[2][24]
Volumes and Storage
Modal provides two storage primitives:
- Modal Volumes: Persistent, distributed storage that can be attached to any Modal function across regions. Built on VolumeFS, volumes support high-throughput reads and writes for model weights, training data, and output artifacts.
- Modal NetworkFileSystem: an older shared, writable file system that Modal's documentation now marks as deprecated and slated for removal.[25]
GPU Availability
Modal provides access to a range of NVIDIA GPUs through a multi-cloud capacity pool that it said spanned more than 20 clouds as of October 2026.[21]
| GPU | Memory | Approximate Hourly Cost (November 2026) |
|---|---|---|
| NVIDIA T4 | 16 GB | $0.59 |
| NVIDIA L4 | 24 GB | $0.80 |
| NVIDIA A10 | 24 GB | $1.10 |
| NVIDIA L40S | 48 GB | $1.95 |
| NVIDIA A100 40 GB | 40 GB | $2.10 |
| NVIDIA A100 80 GB | 80 GB | $2.50 |
| NVIDIA RTX PRO 6000 | 96 GB | $3.03 |
| NVIDIA H100 | 80 GB | $3.95 |
| NVIDIA H200 | 141 GB | $4.54 |
| NVIDIA B200 | 192 GB | $6.25 |
| NVIDIA B300 | 288 GB | $7.10 |
All GPU pricing is billed per second with no minimum commitment.[8] Users can request multiple GPUs per function (e.g., 4x H100 or 8x A100) for multi-GPU training or inference workloads.
Pricing
Modal uses pure usage-based pricing with per-second billing.[8] There are no reserved instances or long-term commitments.[8]
Compute Pricing
| Resource | Price |
|---|---|
| CPU | $0.0000131 per physical core per second |
| Memory | $0.00000222 per GiB per second |
| Sandbox CPU (non-preemptible) | $0.00003942 per physical core per second |
| Sandbox Memory (non-preemptible) | $0.00000667 per GiB per second |
GPU pricing varies by GPU type (see GPU Availability table above).
Plan Tiers
| Feature | Starter (Free) | Team ($250/month) | Enterprise (Custom) |
|---|---|---|---|
| Monthly compute credits | $30 | $100 | Custom |
| Workspace seats | 3 | Unlimited | Unlimited |
| Maximum containers | 100 | 5,000 | Custom |
| GPU concurrency | 10 | 50 | Custom |
| Scheduled functions (crons) | 5 | Unlimited | Unlimited |
| Web endpoints | Limited | Unlimited | Unlimited |
| Log retention | 1 day | 30 days | Custom |
Credit Programs
Modal offers compute credit programs for qualifying organizations:
| Program | Credits Available |
|---|---|
| Startup credits | Tiered free credits for VC-backed startups (amount not published) |
| Academic credits | Up to $10,000 |
Multi-Cloud Infrastructure
Modal operates across multiple cloud providers, including AWS, Google Cloud Platform, and Oracle Cloud Infrastructure, and said in October 2026 that its capacity pool spans more than 20 clouds.[6][21] In September 2024, Oracle announced that Modal was using Oracle Cloud Infrastructure (OCI) bare metal compute instances for inference, fine-tuning, and batch processing workloads.[10] By leveraging OCI's bare metal compute instances, Modal reported improvements in performance for inference, fine-tuning, and batch processing workloads.[10]
The platform's automated resource solver handles multi-cloud scheduling, selecting the most cost-effective GPU capacity across providers based on availability, regional constraints, and pricing.[6] Customers interact only with Modal's unified interface and do not need to manage cloud provider accounts or configurations.
Modal is also available through the AWS and Google Cloud marketplaces, allowing Enterprise customers to apply existing committed cloud spend toward Modal usage.[8]
Security and Compliance
Modal has achieved SOC 2 Type II certification and supports HIPAA-compliant deployments for healthcare and regulated industry use cases. The platform's use of gVisor as a container runtime provides an additional layer of security isolation compared to standard containers, since gVisor intercepts system calls rather than sharing the host kernel directly.[7]
Customers and Use Cases
Modal serves thousands of customers across a range of industries and use cases.[18] Approximately 90% of workloads on the platform are related to AI and ML applications.
| Customer | Use Case |
|---|---|
| Meta | Thousands of concurrent sandboxed environments for reinforcement learning |
| Scale AI | Evaluations, RL environments, and MCP servers with massive volume spikes |
| Substack | Deploying new ML models in hours instead of weeks |
| Ramp | Background coding agents (Inspect) |
| Lovable | AI development platform |
| Quora | AI applications |
| Suno | Music generation |
| OpenPipe | LLM fine-tuning |
| Cognition AI | AI coding agent (Devin) |
| Mistral AI | AI inference |
| Harvey | Legal AI |
| Cartesia AI | Speech AI |
| Allen Institute for AI | Research computing |
Reinforcement learning has become a prominent workload as AI agents scale, with sandboxes serving as a core building block. As Applied Compute CEO Yash Patil put it, "Sandboxes are one of the most important building blocks for Reinforcement Learning."[18]
Common use case categories include:
- Model inference APIs: Deploying LLMs, image generation models, and audio models as scalable endpoints
- Fine-tuning: Training custom models on A100 and H100 GPUs with instant provisioning
- Batch processing: Large-scale data transformation, web scraping, and ETL pipelines
- Computer vision: Image and video processing with YOLO, Tesseract, and other frameworks
- Sandboxed code execution: Running untrusted code from AI agents or end users safely
- Computational biology: Protein folding, molecular modeling, and bioinformatics
Competition
Modal competes in the AI infrastructure and serverless compute market alongside several categories of providers.
| Competitor | Category | Primary Differentiator |
|---|---|---|
| AWS Lambda / SageMaker | Major cloud provider | Comprehensive MLOps platform; deep AWS ecosystem integration; suited for large enterprises |
| Replicate | Serverless inference | One-click model deployment and sharing; strong for demos and prototyping |
| Baseten | Serverless inference | Open-source Truss framework for model packaging; good for early-stage teams |
| Fireworks AI | Inference optimization | Custom inference engine with FireAttention; focused on production throughput and latency optimization |
| Together AI | Full-stack AI cloud | Broad model catalog (200+ models); training and inference; strong fine-tuning support |
| RunPod | GPU cloud | Raw GPU access with serverless and pod-based options; competitive pricing |
| Google Cloud Vertex AI | Major cloud provider | Integrated ML platform within Google Cloud ecosystem |
| CoreWeave | GPU cloud | Large-scale GPU clusters; focused on raw infrastructure rather than developer experience |
Modal differentiates itself through three primary advantages:
- Developer experience: Python-native infrastructure definition with no YAML, Dockerfiles, or cluster management. The decorator-based paradigm allows developers to go from local code to cloud execution with minimal changes.
- Container startup speed: Fast cold starts through custom gVisor-based containers, a FUSE filesystem with content-addressed caching, and CRIU memory snapshotting.[6][7]
- Usage-based pricing: Per-second billing with scale-to-zero means users pay nothing when their code is not running, unlike reserved GPU instances offered by traditional cloud providers.[8]
Open Source Contributions
While Modal itself is a proprietary platform, the company and its founders have contributed to the open-source ecosystem. Erik Bernhardsson created and maintains several open-source projects:
| Project | Description | GitHub Stars |
|---|---|---|
| Luigi | Python workflow scheduler for batch job pipelines (created at Spotify) | 17,000+ |
| Annoy | C++/Python library for approximate nearest neighbor search (created at Spotify) | 13,000+ |
| ANN-Benchmarks | Benchmarking framework for approximate nearest neighbor algorithms | 4,000+ |
Modal's engineering team has also made upstream contributions to gVisor, specifically improving GPU support in the sandboxed container runtime.[7]
Company Culture and Team
Modal's company page lists offices in New York City, Stockholm, San Francisco, and London.[9] The engineering team is notable for its concentration of competitive programming talent, including international olympiad medalists.[9] The team also includes creators of popular open-source projects (Seaborn, Luigi) and experienced engineering leaders from companies like Spotify, Scale AI, and Better.com.
As of October 2026, Modal said it had a team of more than 200 people, and its company page listed open roles across engineering, go-to-market, design, data, and general and administrative functions.[9][21]
Continue reading
- Graphics processing unitA graphics processing unit (GPU) is a processor designed to execute many similar operations in parallel.Mentioned 12 times above
- KubernetesKubernetes, often abbreviated K8s, is an open-source system for deploying and managing containerized applications across a cluster of machines.Mentioned 3 times above
- Machine LearningMachine learning (ML) is a branch of artificial intelligence that gives computers the ability to learn from data and improve their performance on tasks without being explicitly programmed.Mentioned twice above
- Cloud computingCloud computing is the on-demand availability of computer system resources, especially data storage and computing power, delivered over the internet without active management by the end user.Linked in the introduction
- Artificial IntelligenceArtificial intelligence (AI) is the field of computer science dedicated to creating systems that perform tasks normally requiring human intelligence, such as learning, reasoning, recognizing patterns…Linked in the introduction
- NVIDIA B200The NVIDIA B200 is a data center GPU based on the NVIDIA Blackwell microarchitecture, announced by Jensen Huang at NVIDIA's GTC conference in San Jose on March 18, 2024.Links back to this article
More to read
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Bernhardsson, Erik. "What I have been working on: Modal." erikbern.com, December 7, 2022. erikbern.com/...what-ive-been-working-on-modal
- ^1 ^2 ^3 ^4 ^5Modal Labs. "Announcing our $87M Series B." Modal Blog, September 2025. modal.com/...announcing-our-series-b
- ^1 ^2Modal Labs. "Series A Press Release." Modal Blog, October 2023. modal.com/...ailability-and-series-a-press-release
- ^1 ^2TechCrunch. "Modal Labs lands $16M to abstract away big data workload infrastructure." October 10, 2023. techcrunch.com/...big-data-workload-infrastructure
- ^1 ^2 ^3TechCrunch. "AI inference startup Modal Labs in talks to raise at $2.5B valuation, sources say." February 11, 2026. techcrunch.com/...se-at-2-5b-valuation-sources-say
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20Amplify Partners. "How Modal built a data cloud from the ground up." amplifypartners.com/...ta-cloud-from-the-ground-up
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16Modal Labs. "How Modal speeds up container launches in the cloud." Modal Blog. modal.com/...speeding-up-container-launches
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Modal Labs. "Pricing." modal.com/pricing
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Modal Labs. "Company." modal.com/company
- ^1 ^2Oracle. "Modal Uses Oracle Cloud Infrastructure for Large-Scale AI Workloads." September 10, 2024. oracle.com/...-large-scale-ai-workloads-2024-09-10
- ^1 ^2 ^3 ^4 ^5Contrary Research. "Modal Business Breakdown & Founding Story." research.contrary.com/...modal
- SiliconANGLE. "Modal Labs raises $80M to simplify cloud AI infrastructure with programmable building blocks." September 29, 2025. siliconangle.com/...e-programmable-building-blocks
- Built In NYC. "AI-Native Infrastructure Company Modal Raises $87M." October 1, 2025. builtinnyc.com/...m-series-b-1b-valuation-20251001
- BeBeez. "Swedish Modal raises $87M Series B, hits unicorn valuation with AI-native cloud platform." September 30, 2025. bebeez.eu/...luation-with-ai-native-cloud-platform
- ^1 ^2 ^3DZone. "Serverless Cloud Infrastructure With Modal." dzone.com/...modal-python-cloud-platform
- ^Spotify. "Luigi: Python module for complex pipelines of batch jobs." GitHub. github.com/...luigi
- ^1 ^2Spotify. "Annoy: Approximate Nearest Neighbors in C++/Python." GitHub. github.com/...annoy
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Modal Labs. "Modal's Series C: Raising $355M at a $4.65B valuation." Modal Blog, May 21, 2026. modal.com/...modal-series-c
- ^Benzinga. "General Catalyst, Redpoint Fuel Modal's AI Infrastructure Push With $355 Million Raise." May 22, 2026. benzinga.com/...ucture-push-with-355-million-raise
- ^1 ^2Tech Startups. "Modal Labs raises $355M, quadrupling valuation to $4.65B as AI infrastructure demand surges." May 21, 2026. techstartups.com/...i-infrastructure-demand-surges
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Bernhardsson, Erik. "Modal crosses $1B in Annualized Run Rate and raises $750M Series D." Modal Blog, October 14, 2026. modal.com/...1-billion-annualized-run-rate-series-d
- ^1 ^2Bernhardsson, Erik. "About." erikbern.com. erikbern.com/about
- ^Modal Labs. "Modal SDKs for JavaScript and Go." Modal Docs. modal.com/...sdk-javascript-go
- ^Modal Labs. "Modal Notebooks." Modal Docs. modal.com/...notebooks
- ^Modal Labs. "modal.NetworkFileSystem." Modal Docs. modal.com/...modal.NetworkFileSystem
Improve this article
Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.
12 revisions · v12 · 4,166 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Independent fact-check of the full article against its cited sources (150 claims, 34 sources fetched); 12 material + 34 minor fixed
Cite this page: AI Wiki. "Modal (platform)." aiwiki.ai, updated 28 Oct 2026, fact-checked 4 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/modal

