What is Phala Cloud
Phala Cloud provides confidential CPU and GPU infrastructure and private model inference using trusted execution environments with hardware-backed attestation.
Confidential CPU instances
As of August 5, 2026, CPU Confidential VMs have no subscription or minimum and are billed by the minute while running. A tdx.large instance with four vCPUs and 8GB RAM costs US$0.232 per hour. Persistent disk costs US$0.000139 per GB hour, approximately US$0.10 per GB monthly, and continues while the machine is stopped.
GPU On-Demand and Commitment
An H200 with 141GB VRAM costs US$4.80 per hour on demand in the US with a 24-hour minimum. Thirty-day commitments cost US$4.30 hourly in the US or US$4 in India; 180-day commitments cost US$3.80 in the US or US$3.50 in India. Eight-H200 clusters start at US$32 per hour on a 30-day US commitment, while B300 commitments start at US$6.50 per hour for one GPU.
GPU rates include CPU, memory and disk and continue while the reserved GPU machine is stopped. Enterprise adds sustained-use discounts, annual billing, custom support and a 99.9% uptime SLA. Phala also offers token-billed confidential model APIs with live pricing metadata.
Official sources: Phala Cloud pricing, Phala confidential-model pricing metadata.