Hugging Face Inference Endpoints (dedicated) - Hourly instance pricing across AWS and GCP

Fetched 2026-10-04. Documentation page. Hourly rates shown but actual cost calculated by the minute. GPU instances (hourly rate, per instance): AWS: T4 x1 $0.5, T4 x4 $3, L4 x1 $0.8, L4 x4 $3.8, A10G x1 $1, A10G x4 $5, L40S x1 $1.8, L40S x4 $8.3, L40S x8 $23.5, A100 x1 $2.5, A100 x2 $5, A100 x4 $10, A100 x8 $20, H200 x1 $5, H200 x2 $10, H200 x4 $20, H200 x8 $40. GCP: T4 x1 $0.5, L4 x1 $0.7, L4 x4 $3.8, A100 x1 $3.6, A100 x2 $7.2, A100 x4 $14.4, A100 x8 $28.8, H100 x1 $10, H100 x2 $20, H100 x4 $40, H100 x8 $80. No Azure GPU instances listed. CPU instances also available (AWS intel-spr from $0.033/hr, Azure intel-xeon from $0.060/hr, GCP intel-spr from $0.050/hr). Accelerators: AWS Inferentia2 inf2 x1 $0.75/hr, inf2 x12 $12/hr; GCP TPU v5e 1x1 $1.2/hr, 2x2 $4.75/hr, 2x4 $9.5/hr. Pricing formula: instance hourly rate * ((hours * min replicas) + (scale-up hrs * additional replicas)). Requires HF subscription + credit card. Some instance types need quota requests.