LLM Inference & Serverless GPU brand directory

RunPod is an AI-first GPU cloud platform offering on-demand GPU Pods, autoscaling Serverless endpoints, Instant Clusters for distributed compute, and a RunPod Hub marketplace for open-source AI deployment. Its Flash Python SDK further simplifies GPU function deployment via a single decorator. The platform targets the full AI development lifecycle—from experimentation and fine-tuning through to production inference—across a global network of 31 regions.

Together AI is a full-stack AI infrastructure platform — branded as the 'AI Native Cloud' — that enables developers and enterprises to run, fine-tune, and scale open-source AI models in production. It combines a high-performance serverless and dedicated inference API layer, self-service NVIDIA GPU clusters (H100 through GB200), a fine-tuning and model evaluation suite, managed AI-optimized storage, and developer tooling including a code sandbox. The platform is differentiated by an active in-house systems research function that has developed FlashAttention, ATLAS speculative decoding, and ThunderKittens GPU kernels — research improvements that are deployed directly to improve inference throughput and cost efficiency for customers.

Beam is an open-source serverless cloud platform for AI inference, sandboxes, and background jobs. Developers decorate Python or TypeScript functions to run on GPU or CPU-backed containers that launch in under one second, autoscale to thousands of replicas, and bill only for active compute time. The platform supports REST endpoint deployment, async task queues, scheduled cron jobs, sandbox environments with checkpoint/restore for long-running agent sessions, and self-hosting via its open-source runtime (beta9). It is used by startups and Fortune 100 companies to run custom ML models and execute LLM-generated code securely at scale.

Modal is a serverless AI infrastructure platform that turns any Python function into an autoscaling cloud workload with GPU acceleration. Developers decorate Python functions with @app.function(), specify container environments and hardware in code, and invoke workloads via .remote()—Modal handles container builds, scheduling, autoscaling, and logging automatically. Core products include Modal Inference (low-latency LLM and model serving), Modal Training (single- and multi-node GPU fine-tuning), Modal Sandboxes (secure ephemeral environments for AI-generated code execution), Modal Batch (massively parallel batch processing), and Modal Notebooks (collaborative GPU-backed notebooks). The underlying platform includes a custom file system, container runtime, scheduler, and image builder engineered for AI workloads.

Cerebrium is a managed serverless GPU platform for real-time, multimodal AI applications. It allows developers to deploy any AI workload—LLMs, voice pipelines, video models, or custom containers—using a simple CLI or Dockerfile, with automatic autoscaling, per-second billing, and built-in observability across multiple cloud regions.

Baseten is an AI inference platform offering dedicated GPU deployments, pre-optimized Model APIs, multi-node training, and compound AI orchestration. Its proprietary Inference Stack—combining custom model runtimes, multi-cloud GPU management, and developer tooling—enables companies to run open-source and custom AI models in production at high throughput, low latency, and 99.99% uptime across cloud providers.

Sference is an async batch AI inference service running on federated EU spot and preemptible GPU capacity. It delivers up to 75% cost savings versus real-time inference by accepting configurable latency trade-offs, and combines EU data sovereignty, an OpenAI-compatible batch API, BYOM for fine-tuned models, and a compliance runtime (audit trails, DPA, DORA/AI Act readiness) in a single platform aimed at regulated EU SaaS verticals.

Fireworks AI is a frontier AI inference cloud and model lifecycle platform that lets teams run, fine-tune, and scale open-source generative AI models in production. Built by the creators of PyTorch, it combines a high-speed serverless inference API, proprietary GPU optimization (FireAttention), multi-modal model support, and advanced fine-tuning tools—including reinforcement fine-tuning—into a single integrated platform covering the full Build → Tune → Scale workflow.

Lepton AI built a managed AI cloud platform combining a Pythonic developer framework ('Photon') with GPU infrastructure—enabling one-command deployment of LLM inference APIs, distributed training, and HuggingFace model hosting. Acquired by NVIDIA in April 2025, the technology now underpins NVIDIA DGX Cloud Lepton, a multi-cloud GPU compute marketplace connecting developers to tens of thousands of GPUs across a global network of NVIDIA Cloud Partners.

Replicate is a serverless AI model platform that lets developers run, fine-tune, and deploy machine learning models—including 50,000+ community and official models—through a single line of Python or JavaScript code. Its open-source Cog tool standardizes custom model packaging into containers, while its auto-scaling cloud infrastructure handles GPU provisioning, inference serving, model versioning, and billing automatically, with pay-per-second pricing that scales to zero when idle.