You need to enable JavaScript to run this app.
Lake AI Service

Lake AI Service

  • Documentation
    • Lake AI ServiceProduct billingBilling instructions
Copy page
Download PDF
Product billing
Billing instructions
Copy page
Download PDF
Billing instructions

This document introduces the current billing items for the AI Data Lake Service (LAS), as well as the billing methods and logic for each item.

Overview of billing items

After enabling the LAS product, charges will be incurred when using the following LAS functional modules.

warning

  • Starting from LAS AI version 0.11.0, EMR serverless queue resources are used as the 'exclusive queue' for the AI Data Lake Service (this method is recommended), enabling flexible application for development machine creation and task execution, further improving the flexibility and utilization of queue resources. Therefore, beginning with this version, billing for queue resource fees will be generated in the EMR product, and LAS will no longer include queue resource billing items or invoices.
  • When using LAS features, you may need to enable and use other products as required. For example, you may use other storage products (such as TOS or NAS), compute resource products (such as EMR-Serverless), or network products (such as public IPs for VPC). When using these products in conjunction with LAS, the fees incurred by other products will be billed by their respective products and will not appear in the LAS invoice. For details, refer to the billing documentation of the corresponding products.

Feature modules

Billing items

Billing overview

Operator marketplace & operator services

  • ModelArk models
    When processing data with LAS, some operator capabilities are based on ModelArk models. When using such operators, model fees will be charged according to the type of ModelArk model used.
  • Billing method: "Pay-as-you-go", fees are calculated based on the type of model used.

For details, see Large model billing.

  • Self-developed models, model enhancement:
    Some operators use LAS self-developed models, or have enhanced capabilities based on certain models. When using such operators, model fees will be charged according to the type of model associated with the operator.
  • Billing method: "Pay-as-you-go", fees are calculated based on the type of model used.

For details, see Large model billing.

Metadata management

  • Metadata objects:
    To perform unified metadata management of various data using LAS metadata management features, fees for metadata objects will be charged based on the number of metadata objects you create.
  • Billing method: Pay-as-you-go, fees are calculated based on the number of metadata objects.
  • LAS provides a certain free quota. Any usage exceeding the free quota will be charged on a pay-as-you-go basis.

For details, see Metadata objects below.

  • Metadata requests:
    When you create, read, modify, or delete metadata objects for storage management in LAS, corresponding request fees will be charged.
  • Billing method: Pay-as-you-go, charged by number of requests
  • LAS provides a certain free quota. Usage exceeding the free quota is billed on a pay-as-you-go basis.

For details, see Metadata requests below.

Data development

  • Cloud disk:
    To develop data tasks using the LAS development machine, you need to mount a cloud disk. The cloud disk is mounted as persistent storage at the root directory (/), and data in this directory will be retained after shutdown or restart.
  • Billing method: "Pay-as-you-go"
  • Billed based on cloud disk capacity (GiB/hour). The billing details vary depending on the resource specifications of the cloud disk.

For details, see Billing details: Cloud disk below.

Billing details: Operators/models

For details, see Large model billing.

Billing details: Metadata management

The metadata management feature of LAS only supports pay-as-you-go billing, that is, postpaid billing.

Metadata objects

Definition of billing items

  • Metadata objects refer to all levels of metadata stored and managed in LAS, including all databases, tables, volumes, functions, partitions, and indexes.
  • Relationship between datasets and metadata
    • General datasets: When metadata storage is specified for a general dataset (for example, when creating or editing a general dataset, a metadata table is associated), the corresponding general dataset is considered a metadata object.
    • Inference datasets: Each inference result set is backed by a metadata table. Depending on the dataset type, text data uses an iceberg table, and multimodal data uses a lance table. Each iceberg table or lance table is a metadata object.

Billing details

After billing begins, both existing and newly added metadata will be charged.

Billing unit

Billing mode

Applicable region

Billing method

Number of metadata objects

Pay-as-you-go

  • North China 2 (Beijing)
  • East China 2 (Shanghai)
  • South China 1 (Guangzhou)
  • North China 2 (Beijing) Autonomous Driving Cloud
  • Free quota: First 1,000,000 per month are free
  • Billed portion: For usage exceeding 1,000,000, each additional 10,000 is charged 0.5 RMB/month
  • Asia Pacific Southeast (Johor)
  • Free quota: First 1,000,000 per month are free
  • Billed portion: For usage exceeding 1,000,000, each additional 10,000 is charged 0.8 RMB/month

Billing/settlement cycle

The number of objects is counted monthly. On the 1st of each month, billing and settlement are performed.

Metadata requests

Definition of billing items

  • Metadata access refers to the actions of creating, reading, modifying, and deleting metadata objects stored and managed in LAS.
  • Metadata object access can be performed by executing DDL statements in one or more compute engines integrated with LAS (such as EMR Serverless Spark, Hive on EMR, SparkSQL on EMR, Serverless Presto, LAS, and so on), or directly operating metadata objects via the LAS product console or API.

Billing details

Billing unit

Billing mode

Applicable region

Billing method

Number of metadata requests

Pay-as-you-go

  • North China 2 (Beijing)
  • East China 2 (Shanghai)
  • South China 1 (Guangzhou)
  • North China 2 (Beijing) Autonomous Driving Cloud
  • Free quota: First 1,000,000 requests per month are free
  • Billed portion: For usage exceeding 1,000,000, each additional 10,000 requests is charged 0.05 RMB/month
  • Asia Pacific Southeast (Johor)
  • Free quota: First 1,000,000 requests per month are free
  • Billed portion: For usage exceeding 1,000,000, each additional 10,000 requests is charged 0.08 RMB/month

Billing/settlement cycle

The number of objects is counted monthly. On the 1st of each month, billing and settlement are performed.

Billing details: Cloud disk

Billing details

Billing unit

Billing mode

Applicable region

Unit price

Cost calculation

Cloud storage disk

Pay-as-you-go

  • North China 2 (Beijing)
  • East China 2 (Shanghai)
  • Extreme SSD PL0 unit price: 0.00105 RMB/GiB/hour
  • Extreme SSD FlexPL unit price: 0.0021 RMB/GiB/hour

Cost = cloud storage disk unit price × capacity × billing duration

  • Asia Pacific Southeast (Johor)
  • Extreme SSD PL0 unit price: 0.00165 RMB/GiB/hour
  • Extreme SSD FlexPL unit price: 0.0033 RMB/GiB/hour

Billing/settlement cycle

The billing cycle is accurate to the second.

Billing details: queue resources (for legacy users)

warning

  • Starting from the LAS AI version released on January 8, 2026, EMR serverless queue resources are used as "dedicated queues" for AI data lake services (this method is recommended), allowing flexible application for development machine creation and task execution, further improving the flexibility and utilization of queue resources. Therefore, from this version onward, the billing for queue resource fees will be generated within the EMR product, and LAS will no longer include queue resource billing items or fee invoices.

Compute queue

Billing unit

Applicable region

Queue type

Unit price

Billing mode

Compute queue

North China 2 (Beijing), East China 2 (Shanghai), South China 1 (Guangzhou)

Public queue

  • CPU: 0.35 RMB/CU/hour
  • GPU: 4.55 RMB/GU/hour

Pay-as-you-go

Dedicated queue

  • Dedicated CPU queue:
    • Compute-optimized (1:2): 135 RMB/CU/month
    • Standard (1:4): 160 RMB/CU/month
    • Memory-optimized (1:8): 225 RMB/CU/month
  • Dedicated GPU queue:
    • L20*1 22vCPU 120GiB Mem 48G VRAM: 12,800 RMB/node/month
    • A10*1 28vCPU 112GiB Mem 24G VRAM: 6,560 RMB/node/month
    • A30*1 28vCPU 234GiB Mem 24G VRAM: 13,120 RMB/node/month
    • A100*1 14vCPU 245GiB Mem 80G VRAM: 32,960 RMB/node/month
    • V100*1 8vCPU 32GiB Mem 32G VRAM: 8,000 RMB/node/month
    • T4*1 8vCPU 26GiB Mem 16G VRAM: 4,480 RMB/node/month
    • L4*1 12vCPU 64GiB Mem 24G VRAM: 6,240 RMB/node/month
    • 4090D*1 22vCPU 90 GiB Mem 24G VRAM: 7,680 RMB/node/month
    • V100*2 16vCPU 64 GiB Mem 64G VRAM: 16,000 RMB/node/month

Annual/monthly subscription

Asia Pacific Southeast (Johor)

Public queue

  • CPU: 0.60 RMB/CU/hour
  • GPU: 8 RMB/GU/hour

Pay-as-you-go

Dedicated queue

  • Dedicated CPU queue:
    • Compute-optimized (1:2): 235 RMB/CU/month
    • Standard (1:4): 260 RMB/CU/month
    • Memory-optimized (1:8): 360 RMB/CU/month
  • Dedicated GPU queue:
    • L20*1 22vCPU 120GiB Mem 48G VRAM: 20,800 RMB/node/month

Annual/monthly subscription

General queue

Resources purchased for general queues are used to create development machines and deploy online services, enabling rapid setup of data processing environments and services. Billing is based on the following methods:

Billing unit

Applicable region

Unit price

Billing mode

General queue

  • North China 2 (Beijing)
  • East China 2 (Shanghai)
  • CPU：
    • Standard (1:4): 0.35 RMB/CU/hour
    • Memory-optimized (1:8): 0.5 RMB/CU/hour
  • GPU：
    • A10*1 28vCPU 112GiB Mem 24G VRAM: 4.55 RMB/GPU/hour
    • L20*1 22vCPU 120GiB Mem 48G VRAM: 17.5 RMB/GPU/hour
    • L4*1 12vCPU 64GiB Mem 24G VRAM: 8.06 RMB/GPU/hour
    • T4*1 8vCPU 26GiB Mem 16G VRAM: 7 RMB/GPU/hour
    • V100*1 8vCPU 32GiB Mem 32G VRAM: 14.7 RMB/GPU/hour
    • A100*1 14vCPU 245GiB Mem 80G VRAM: 50.4 RMB/GPU/hour
    • A30*1 28vCPU 234GiB Mem 24G VRAM: 8.05 RMB/GPU/hour
    • 4090D*1 22vCPU 90 GiB Mem 24G VRAM: 8.75 RMB/GPU/hour

Pay-as-you-go

  • CPU：
    • Standard (1:4): 160 RMB/CU/month
    • Memory-optimized (1:8): 225 RMB/CU/month
  • GPU：
    • L20*1 22vCPU 120GiB Mem 48G VRAM: 12,800 RMB/node/month
    • A10*1 28vCPU 112GiB Mem 24G VRAM: 6,560 RMB/node/month
    • A30*1 28vCPU 234GiB Mem 24G VRAM: 13,120 RMB/node/month
    • A100*1 14vCPU 245GiB Mem 80G VRAM: 32,960 RMB/node/month
    • V100*1 8vCPU 32GiB Mem 32G VRAM: 8,000 RMB/node/month
    • T4*1 8vCPU 26GiB Mem 16G VRAM: 4,480 RMB/node/month
    • L4*1 12 vCPU 64 GiB Mem 24G GPU memory: 6,240 RMB/server/month
    • 4090D*1 22 vCPU 90 GiB Mem 24G GPU memory: 7,680 RMB/server/month

Annual/monthly subscription

Southeast Asia (Johor)

  • CPU：
    • Compute-optimized (1:2): 0.55 RMB/CU/hour
    • Standard (1:4): 0.6 RMB/CU/hour
    • Memory-optimized (1:8): 0.85 RMB/CU/hour
  • GPU：
    • A10*1 28 vCPU 112 GiB Mem 24G GPU memory: 7.8 RMB/GPU/hour
    • L20*1 22 vCPU 120 GiB Mem 48G GPU memory: 30 RMB/GPU/hour

Pay-as-you-go

  • CPU：
    • Compute-optimized (1:2): 235 RMB/CU/month
    • Standard (1:4): 260 RMB/CU/month
    • Memory-optimized (1:8): 360 RMB/CU/month
  • GPU：
    • L20*1 22 vCPU 120 GiB Mem 48G GPU memory: 20,800 RMB/server/month

Annual/monthly subscription

Related documentation
Last updated: 2026.05.22 14:59:17