Phala Cloud Pricing

Last updated: July 3, 2026

Phala Cloud has three products, each priced in its own way:

• CPU Instances: private machines billed by the minute, like AWS or GCP.
• GPU Rental: whole GPUs set aside for you, rented by the hour with a minimum period.
• LLM API: private AI inference. Models and pricing are in the Confidential AI docs.

CPU Instances

Prices are shown per hour. We charge only for the minutes your machine runs. There is no subscription and no minimum.

You deploy with docker compose, but you get a whole private machine (a CVM). The price is for the machine, not for each container.

Each machine has two charges: compute and storage.

Compute

The compute price depends on the instance type you pick. It is charged only while the machine runs. When you stop the machine, the compute charge stops the moment we receive your stop request. Shutting down then takes a little while, because a TEE machine must wipe its memory. That shutdown time is not billed.

Storage

Storage costs $0.000139 per GB per hour, which is about $0.10 per GB per month. It is charged while the machine runs and also while it is stopped, because your disk data is kept. To stop the storage charge, delete the machine.

• The smallest disk is 20 GB.
• You get the full disk size for your data. System files do not take space from it.

CPU Pricing Example

A machine using tdx.large (4 vCPU, 8 GB RAM) with a 100 GB disk:

While running:

• Compute: $0.232 per hour
• Storage: 100 × $0.000139 = $0.0139 per hour
• Total: about $0.246 per hour

While stopped:

• Compute: $0
• Storage: 100 × $0.000139 = $0.0139 per hour
• Total: about $0.014 per hour

What You Don't Pay For

• Stopped machines: compute is $0 while a machine is off. Only storage is charged.
• Setup time: pulling images and creating the machine is free.
• Shutdown time: billing stops when you send the stop request, so memory wiping is free.
• Failed starts: if a machine fails to start, you are not charged.

Account Tiers and Limits

Tier limits apply to CPU machines only. They do not limit GPU rentals or the LLM API.

New accounts start at Tier 1. Your account moves up to Tier 2 automatically based on how much you have spent. For Enterprise, contact us at [email protected].

GPU Rental

GPU machines are rented through GPU TEE for AI and machine learning workloads.

When you rent a GPU, the whole GPU is set aside for you. No one else can use it, even while your machine is stopped. So the clock runs on the GPU itself, not on your usage. This is why a GPU rental cannot be paused.

One hourly rate covers everything: the GPU, CPU, memory, and disk. There are no separate charges. There are two plans: On-Demand and Commitment.

On-Demand

You pay for the minimum rental period when you place the order. This first payment is not refundable. After the minimum period, your machine keeps running and we charge your account hour by hour. You can end the rental at any time after that.

For example: if the rate is $2.50 per hour and the minimum period is 24 hours, you pay $60 when you order. After 24 hours, you pay $2.50 for each extra hour until you end the rental.

Commitment

You pay for the full term (30, 180, or 365 days) when you place the order. In return, the hourly rate is lower than On-Demand. The payment is not refunded, even if you stop using the GPU early.

Can I stop a GPU machine to save money?

No. The GPU stays set aside for you while the machine is stopped, so stopping does not stop the cost.

• On-Demand: stopping the machine ends the rental and deletes the machine. The disk is deleted too, so your data is not kept. Back up anything you need before you stop. If you need a GPU again later, place a new order.
• Commitment: you can stop and start the machine, but the term is already paid. The GPU stays yours until the term ends.

What if my machine gets stuck in an error state?

The GPU is still set aside for you, so charges continue. To stop the charges, delete the machine. This ends the rental. If you think something is wrong on our side, contact us at [email protected].

Do credits work for GPU rental?

Yes. We use your free credits first, then your balance.

Do I get money back if I end a rental early?

No. The minimum period (On-Demand) or the full term (Commitment) is paid up front and is not returned.

Current Rates

Rates depend on the GPU model and the region. The rates below are live.

Network and Regions

• CPU machines run in US East. Europe is not open by default; contact support to enable it.
• GPU machines are only in the regions shown in the rates table above. There are no GPU machines in Europe.
• Need a different region? Open a support ticket.
• All prices are in USD.
• Each machine includes bandwidth based on its size.

Legacy Free CVM Instance

For legacy accounts with existing free CVM instances:

• Legacy instances are automatically reclaimed 6 months after their creation date.
• Inactive or non-compliant instances may be deleted at any time.
• Move to a paid plan to keep your service running.

Enterprise Support

For larger workloads we offer:

• Volume discounts for sustained usage
• Annual billing options
• Custom support packages
• 99.9% uptime SLA

Questions? Contact us at [email protected] for enterprise pricing.