Big Data & AI Product Monthly Newsletter- Feb & Mar 2026

Big Data & AI Product Monthly Report [Feb & Mar 2026], covering tech highlights, product and feature releases, market updates, and customer application practices, helping you stay up to date with the latest from Alibaba Cloud Big Data & AI.

Product Releases

PAI - DSW Integrated with OpenClaw: Intelligent Cloud Development

PAI-DSW provides an automated OpenClaw installation solution, supporting system control, persistent memory, and scheduled push capabilities. Users can interact via Web UI or DingTalk, letting AI agents "live next to compute power."

PAI - PAI-EAS PD Separation Smart Configuration Release

When users deploy large-scale distributed inference services based on PD separation, the PAI assistant intelligently recommends appropriate PD ratios and parallelism strategies based on model characteristics, machine configurations, and I/O length, improving resource utilization and end-to-end performance.

PAI - Global Search Capability Release

PAI global search supports one-click retrieval of DSW instances, DLC tasks, EAS services, models, and other resources. Quickly locate target assets via keywords to improve resource search efficiency.

PAI - PAI-EAS Model Pre-warm Cache Feature Release

The model pre-warm cache service is an independent service for preloading specified model caches, providing a high-speed access data source for inference services with model cache acceleration enabled. Suitable for scenarios requiring OSS/NAS-mounted large model files such as LLMs, AI image generation, and AI video generation.

MaxCompute - Iceberg External Table Support

MaxCompute supports users submitting DDL for a specified Schema to build Iceberg external tables, and also supports dynamically generating table Schema and partition information based on metadata in Iceberg files.

MaxCompute - MaxFrame New Multimodal Data Processing Operator Module

MaxFrame now includes built-in Image operators, supporting distributed processing of massive image data.

MaxCompute - MaxFrame AI Function

MaxFrame AI Function receives a major upgrade, adding support for Alibaba Cloud Bailian commercial large models (qwen3-max, text-embedding-v4), and providing a per-Token billing model (pay-as-you-go).

MaxCompute - MaxFrame New Local Debugging

MaxFrame adds a Local Debug Mode. When enabled, MaxFrame UDF functions can be executed directly in the local Python environment, supporting IDE breakpoint debugging, fully available offline.

MaxCompute - MaxFrame OSS Data Mounting Enhancement

MaxFrame enhances OSS data mounting, now supporting multiple OSS Bucket mounting and path prefix configuration.

Hologres - AI Function Supports Model Studio

Hologres is deeply integrated with Alibaba Cloud Bailian. Users can deploy Bailian models in Hologres via API Key and then call them using AI Functions. Data stays in-database for AI development and application building.

Hologres — Long-Term Memory for LLMs with Mem0

Give AI assistants persistent memory across sessions. Hologres leverages its high-performance vector engine and the open-source Mem0 framework to automatically extract, store, and retrieve conversation history, user preferences, and key facts

Hologres - Dynamic Table Supports Automatic Query Rewriting

Dynamic Table supports automatic query rewriting: when pre-aggregation has been done on a base table via Dynamic Table, the optimizer can automatically rewrite base-table aggregate queries to target the DT, improving query efficiency.

Hologres - User Id Encoding Function Support

By mapping high-cardinality strings to integer IDs, combined with Dynamic Table incremental computation, this enables ultra-low-cost, ultra-fast real-time UV computation at massive scale.

Hologres - EXTERNAL_FILES Function for Lake Data Analysis Without External Tables

Adds the EXTERNAL_FILES function, supporting direct analysis of Parquet or ORC format files on OSS via standard SQL without creating external tables.

Hologres - Zone-Redundant Storage (3AZ) Supports Japan and Hangzhou Regions

Target customers: Financial, internet, and other customers with disaster recovery needs. New feature/spec: Zone-Redundant Storage (3AZ) now supports Japan and Hangzhou regions.

Hologres - Dify Supports Dual-Path Retrieval for RAG Using Hologres

Dify 1.13.1 supports using Hologres full-text + vector retrieval to achieve dual-path retrieval for RAG.

DataWorks - Data Operations Agent Launch

The Data Operations Agent can automatically generate structured diagnostic reports by integrating multi-dimensional analysis including dependency chains, resource utilization, historical run trends, change impact, log anomalies, and data quality.

DataWorks - Workspace Upgrade from Simple Mode to Standard Mode

Supports upgrading workspaces from simple mode to standard mode, including workspaces with the new DataStudio IDE enabled, helping teams achieve more refined permission management and task collaboration.

DataWorks - Data Map Adds DLF Table Permission Requests and Data Preview

In Data Map, users can request permissions for DLF tables. A new data preview feature based on Serverless Spark compute resources and custom scripts improves data acquisition efficiency and data usage security.

DataWorks - ChatBI Supports Deep Analysis Mode

DataWorks ChatBI adds a deep analysis mode, supporting multi-dimensional drill-down, root cause analysis, time-series comparison, anomaly detection, and intelligent attribution. It automatically uncovers trends, correlations, and underlying causes in data, enhancing analysis and decision-making capabilities.

DataWorks - Data Service Supports Cloud-Native API Gateway

DataWorks Data Service adds support for cloud-native API gateways. Users can quickly encapsulate data as APIs via wizard or script mode, with one-click publishing providing high-availability, monitorable online calls. This enables service-oriented data output, accelerating business integration and innovation.

Realtime Compute for Apache Flink — AI-Driven Real-Time Analytics

Pipe live streaming data through LLMs hosted in Alibaba Cloud Model Studio for real-time sentiment analysis, intent recognition, content moderation, and semantic search — all at millisecond latency within the Flink pipeline. No separate inference service needed.

EMR Serverless StarRocks — AI Center (Beta)

11 built-in AI SQL functions — sentiment analysis, auto-tagging, PII masking, summarization, translation, semantic ranking, and custom model calls — callable directly inside queries. Free trial: 1 million tokens per user.

EMR Serverless Spark — First Cloud Support for Apache Spark 4.0

Among the first cloud providers to support Apache Spark 4.0, with upgrades across SQL performance, Python runtime, and semi-structured data processing. New features include NAS-managed file directories, DLF catalog integration for native Paimon/Iceberg support, subscription billing at queue level, and significantly improved PySpark Notebook experience.

Vector Retrieval Service for Milvus — Multimodal Search

Four retrieval modes out of the box: text-to-image, image-to-image, text-to-text, and image-to-text — powered by Qwen-VL vision-language model and DashScope embedding API.

Elasticsearch - Cross-Region High Availability Solution

Based on Cross-Cluster Replication (CCR), achieves primary-standby disaster recovery high availability within or across regions.

Elasticsearch - Agent Memory Solution

Provides an Agent memory storage and management solution based on ES + mem0, helping AI applications build long-term memory capabilities.

Case Studies

Foundational Model company MiniMax adopted DataWorks, MaxCompute, Hologres, and MaxFrame, cutting compute costs by 75%, lowering storage costs by 40%, and reducing operational overhead by 50%.

Gaming studio Hypergryph adopted EMR Serverless Spark and StarRocks, cutting compute time by 50% and accelerating SLA delivery by 1.5 hours.