This topic describes the billing rules, overdue payment policy, and renewal policy for PAI-Rec.
Billing items
|
Billing item |
Description |
Billing method |
|
PAI-Rec instance |
PAI-Rec charges by instance. On top of a standard instance, you can also purchase the following two value-added features:
|
|
|
Associated cloud services |
To build a complete recommendation system, you need not only a PAI-Rec instance but also other cloud services such as MaxCompute, PAI-EAS, Hologres, OSS, PAI, and Flink. Resource consumption for these services is billed according to their respective pricing models. Note that PAI-TF jobs can use subscription-based MaxCompute resources. If you run PAI-TF jobs on a pay-as-you-go basis, they are billed under the PAI platform pricing model. PAI platform algorithm binning is billed under data_analysis billing. pai_etrec and swing are billed under the default algorithm type. |
For billing details, see:
|
|
Solution implementation or advanced capability deployment (optional) |
If you need Alibaba Cloud engineers to help with system setup or algorithm customization, contact your account manager to discuss collaboration terms. After project completion, you must pay a fee based on the estimated person-days required for customization. |
Contact your account manager for details. |
Billing method
PAI-Rec uses a subscription (prepaid) billing model. Estimate your workload and pay upfront for a PAI-Rec instance with the appropriate specification. You pay first and then use the service. The subscription is valid for one year from the purchase date. Any unused portion after expiration automatically becomes invalid. You can purchase additional capacity as needed.
Pricing varies by region. Recommendation engines are priced differently across regions.
|
Area |
Region |
Standard Edition price |
Custom recommendation solution (value-added feature) |
Operation tools (value-added feature) |
|
Asia-Pacific |
|
USD 752/month |
USD 301/month |
USD 151/month |
|
USD 1,204/month |
|||
|
Europe and Americas |
|
Overdue payments
Reason for overdue status
Your renewal account has an insufficient balance, and your account is now overdue.
Service suspension due to overdue payment
If a dedicated resource group expires without successful renewal, its machines are released. After release, the machine status changes to stopped, and any PAI-Rec service using this resource group changes to pending.
-
If you renew successfully within 15 calendar days (15 × 24 = 360 hours) from the expiration date, the system automatically restores the stopped machines.
-
If you do not renew within 15 calendar days from the expiration date, the system permanently deletes the stopped machines after the 15-day period ends.
To avoid service limitations during your PAI-Rec instance term—including usage restrictions on cloud resources such as MaxCompute and PAI—maintain a sufficient balance in your Alibaba Cloud account.
Check overdue amount
Go to the Expenses and Costs homepage to view your overdue amount.
In the Overview section at the top of the page, check your Available credit, Billable amount, and Total outstanding balance (your overdue amount).
Renewal policy
You can renew using one of the following methods:
-
Auto-renewal
To avoid manual renewal each time, select Auto-renewal when you purchase a PAI-Rec instance.
Note-
The first deduction occurs at 08:00 on the 15th day before expiration. Ensure your payment method has sufficient funds. Auto-renewal attempts occur once daily from 15 days before expiration until the day before expiration. If your instance expires tomorrow, choose manual renewal instead.
-
If you manually renew before the auto-deduction date, the system performs auto-renewal at the next expiration cycle.
-
-
Manual renewal
Before expiration or within 15 days after expiration, log on to the PAI-Rec Management Console. On the Basic information page, select your target instance and click Renew on the right.
Appendix: Estimated costs for PAI-Rec (Standard Edition instance + value-added features)
These estimates are for reference only. Actual charges appear on your invoice.
Recommendation solution complexity causes significant cost variation. Key factors include the number of users and items, use of vector recall, cold-start algorithms for items, complex ranking models, and online learning. To reduce costs, consider these measures: use PAI-EAS instances with automatic scaling, use subscription-based MaxCompute, regularly clean unused data in Hologres or OSS, and use incremental training instead of full training.
Based on DAU (Daily Active Users), the estimated resource consumption for the complete recommendation system—including offline training and online serving—is as follows (for reference only):
|
Business scale |
Estimated resource consumption (list price/month) |
|
DAU under 50,000 |
USD 3,000–6,000 |
|
DAU 50,000–100,000 |
USD 5,300–10,500 |
|
DAU 100,000–200,000 |
USD 10,500–22,600 |
|
DAU 200,000–500,000 |
USD 22,600–45,000 |
|
DAU 500,000–2,000,000 |
USD 52,600–105,000 |