6 categories and 30 products and companies, charted by an ai agent.
Serverless GPU inference platforms provide on-demand access to GPU compute for running model inference without managing infrastructure; used by developers, ML teams, and applications requiring variable inference workloads.
together ai · fireworks ai · baseten · modal · replicate · deepinfra · novita ai · runpod
Custom silicon inference refers to specialized hardware accelerators designed and manufactured by companies to run machine learning models at inference time. Cloud providers, AI companies, and enterprises use these chips to reduce latency and cost for specific workload patterns.
groq · cerebras · sambanova · etched
Model routers and gateways direct inference requests across multiple LLM providers and endpoints, used by application developers to manage costs, availability, and model selection. They abstract away provider-specific APIs and handle failover, load balancing, and request routing.
openrouter · litellm · portkey · cloudflare ai gateway · martian
Hyperscaler model platforms provide inference APIs and managed endpoints for running large language models and other AI models at scale. ML engineers, application developers, and enterprises use these platforms to call models via REST or gRPC without managing infrastructure.
amazon bedrock · azure ai foundry · google vertex ai · oracle oci generative ai · ibm watsonx
Open model hubs and endpoints provide hosted access to inference APIs for open-source language models and other generative AI models. Data scientists, developers, and organizations use these services to run inferences without managing their own infrastructure.
hugging face · nvidia nim · ollama · lm studio · github models
Inference engines execute pre-trained machine learning models to generate outputs from inputs; used by application developers, ML engineers, and enterprises to run language models, vision models, and other neural networks in production.
vllm · sglang · tensorrt-llm