Modal AI
Modal is a serverless platform that runs CPU, GPU, and data-intensive compute at scale with sub-second cold starts and instant autoscaling. Designed for AI and data teams, it enables developers to deploy inference, training, batch processing, and sandboxed environments by writing code directly—without YAML or infrastructure configuration. The platform includes elastic GPU scaling across multiple clouds, integrated observability, and a globally distributed storage layer optimized for fast model loading and high-throughput workloads.
Modal is a serverless platform purpose-built for AI and data teams to deploy inference, training, batch processing, and sandboxed environments without managing infrastructure. Developers write code directly in Python—no YAML, Dockerfiles, or configuration files required—and Modal handles deployment, scaling, and observability automatically. The platform delivers sub-second cold starts and instant autoscaling across CPU, GPU, and data-intensive workloads, with elastic GPU capacity spanning thousands of GPUs across multiple clouds and the ability to scale back to zero when idle. Its AI-native runtime is engineered for fast model initialization and container startup (100x faster than Docker), while a globally distributed storage layer optimizes model loading and high-throughput data access. Modal includes unified observability with integrated logging, first-party integrations for cloud buckets and MLOps tools, multi-cloud capacity scheduling, and security features including SOC2, HIPAA compliance, and data residency controls. The platform powers production workloads for companies like Scale, Mistral, Suno, Meta, and You.com across use cases including LLM inference, audio transcription, coding agents, and image and video generation.
Modal may use aggregated de-identified data for improving and marketing products, but explicitly excludes customer-uploa
No training opt-out mechanism was identified in the legal documents.
Moderate risk. Suitable for internal data but avoid confidential information.
Modal may use aggregated de-identified data for improving and marketing products, but explicitly excludes customer-uploaded data. However, the broad permission to use aggregated data for 'business purposes' including product improvement creates moderate risk for shadow AI training.
“Modal may freely use Aggregated De-identified Data for Modal's business purposes (including without limitation, for purposes of improving, testing, operating, promoting and marketing Modal's current and future products and services).”
Customer grants Modal a 'limited license' to use customer data only to provide the service. However, customer feedback receives a broad, perpetual, irrevocable, worldwide, transferable, sublicensable license, creating moderate licensing risk.
“Customer hereby grants to Modal a nonexclusive, worldwide, perpetual, irrevocable, transferable, sublicensable, royalty-free, fully paid up license to use and exploit any Feedback for any purpose.”
Third-party service providers have access to personal information and are obligated not to disclose it, but the agreement doesn't explicitly restrict sharing of aggregated data or specify which third parties have access. Moderate risk.
“We may employ third-party companies and individuals...these third parties have access to your Personal Information...they are obligated not to disclose or use the information for any other purpose.”
Customer account will be deleted within 60 days of termination, but no explicit data deletion policy during active use. Aggregated de-identified data can be retained indefinitely. Moderate risk due to lack of granular deletion controls.
“Customer's Service account will be deleted within 60 days following the termination date.”
Customer retains all right, title and interest in customer data and owns all customer-generated content. Modal retains ownership of the service itself. Clear IP ownership allocation with low risk.
“Customer shall retain all right, title and interest in and to the Customer Data, including all intellectual property rights therein. As between Modal and Customer, Customer owns all Customer Data.”
Agreement references a Data Processing Addendum and security documentation available on request, but no specific mention of SOC2, GDPR, HIPAA, or ISO certifications in the provided text. Moderate risk due to lack of explicit compliance certifications.
“Modal's Data Processing Addendum found at https://modal.com/legal/dpa which is incorporated herein by reference.”
Track and measure your AI tools with Olakai's observability platform.
Explore Olakai