Cerebrium

Cerebrium is a serverless AI infrastructure platform for machine learning engineers and teams building and deploying GPU-backed AI applications, including real-time voice, video, and LLM inference.

Key capabilities include:

• Serverless GPUs with 2–4 second cold starts and 12+ GPU types (including A100 and H100) for matching hardware to workload needs
• Autoscaling from 1 to 10,000+ concurrent requests, plus support for batching and asynchronous jobs for background workloads
• Multiple serving options, including REST, WebSocket, and streaming endpoints; designed for low-latency execution (under 35 ms overhead)
• Developer workflow via a Python library and CLI (one-line deploy), plus a web dashboard; includes orchestration, versioning, and observability support such as OpenTelemetry

Pricing: Pay-per-use (billed per second or millisecond of inference time); includes a free account with GPU credits (amount varies by offer).

Notable for combining low cold starts, broad GPU selection, and multi-region deployments (5 regions) for teams that want to ship model endpoints without managing Kubernetes or long-running GPU instances. It also lists compliance support (including SOC 2, HIPAA, and GDPR) and is available on AWS Marketplace.