AI Hypercomputer
Feature
Generally available: The AI Hypercomputer documentation includes support for the A3 Mega and A3 High machine types. The addition of A3 Mega and A3 High expand the available compute options for training and serving large-scale AI models on Google Kubernetes Engine (GKE) and Compute Engine.
To support this effort, several new pages have been added and existing pages updated in the AI Hypercomputer documentation where relevant. These updates provide comprehensive guidance for deploying, managing, and optimizing the A3 Mega and A3 High machine types within the AI Hypercomputer stack.
New documentation pages include the following:
- Create an AI-optimized instance with A3 High or A3 Mega
- Create AI-optimized instances in bulk with A3 High or A3 Mega
- Create GKE Autopilot clusters which use A3 Mega or A3 High
- Create GKE Standard clusters which use A3 Mega or A3 High
- Run NCCL on custom GKE clusters that use A3 Mega and A3 High
AlloyDB for PostgreSQL
Announcement
The following vector search improvements are now available in Preview:
- AlloyDB now supports Vector assist. Vector assist is an AlloyDB extension that simplifies the deployment and management of your AlloyDB vector workloads. It helps you set up production-ready vector search capabilities, such as embedding generation, query optimization, and index creation for vector types like HNSW. For more information about vector assist, how it works, and its limitations, see Vector assist overview.
- You can now defer ScaNN index creation on an empty table or a table with insufficient rows until the table has sufficient data. For more information, see Create a ScaNN index.
- The
alloydb_scannextension now supports four-level tree indexes, providing support for tables with up to 10 billion vector rows. For more information, see Four-level ScaNN tree indexes.
Feature
Adaptive filtering from inline filtering to pre-filtering is now generally available (GA). With AlloyDB AI, you can use adaptive filtering to optimize filtered vector searches. This feature enables the query optimizer to use cost-based analysis to dynamically choose the most efficient filtering strategy—either inline filtering or pre-filtering—based on real-time data distributions. This improves filtered vector search performance without requiring manual tuning or intervention. Note that the feature adaptive filtering from pre-filtering to inline filtering is still in Preview.
For more information, see Understand adaptive filtering in AlloyDB AI.
Announcement
The alloydb_scann extension is updated to include the following
vector search improvements. These features are generally available
(GA):
- By default, new ScaNN vector index builds are automatically tuned. Manually-tuned indexes can be converted to automatically-tuned indexes. For more information, see Create a ScaNN index.
- You can now automatically maintain your ScaNN vector indexes. AlloyDB incrementally manages your index such that when your dataset grows, AlloyDB updates centroids and splits large outlier partitions to provide better QPS and search results. For more information, see Maintain indexes automatically.
BigQuery
Feature
Conversational analytics now supports querying Lakehouse tables that connect to the Apache Iceberg REST catalog or are federated to an external catalog. For more information, see Query BigLake data with natural language.
This feature is in Preview.
Feature
You can now use Colab Data Apps to transform your data analyses from Colab notebooks into polished, interactive applications.
This feature is in Preview.
Feature
You can now use the
AI.KEY_DRIVERS function
to identify segments of data that cause statistically significant changes to a
summable metric.
This feature is in Preview.
Bigtable
Feature
You can stream messages from Pub/Sub directly to a Bigtable table using Bigtable subscriptions. This feature lets you write streaming messages to Bigtable without needing a separate subscriber such as Dataflow. This feature is available in Preview.
Cloud Run
Feature
Support for specifying custom CPU or concurrency targets using scaling controls is in Preview.
Cloud SQL for MySQL
Feature
The Cloud SQL remote MCP server is generally available (GA). The Cloud SQL remote MCP server lets you interact easily with Cloud SQL instances from LLMs, AI applications, and AI-enabled development platforms.
Cloud SQL for PostgreSQL
Feature
The Cloud SQL remote MCP server is generally available (GA). The Cloud SQL remote MCP server lets you interact easily with Cloud SQL instances from LLMs, AI applications, and AI-enabled development platforms.
Cloud SQL for SQL Server
Feature
The Cloud SQL remote MCP server is generally available (GA). The Cloud SQL remote MCP server lets you interact easily with Cloud SQL instances from LLMs, AI applications, and AI-enabled development platforms.
Compute Engine
Feature
Generally available: To ensure data consistency when backing up multiple disks, you can use consistency groups of instant snapshots to back up a group of disks at the same point in time.
For more information, see About instant snapshots.
Feature
Generally available: You can rotate the customer-managed encryption key (CMEK) used to encrypt a disk, standard snapshot, or archive snapshot to a new key version without downtime.
Generally available: You can change the CMEK used to encrypt a disk, standard snapshot, or archive snapshot to a different key without downtime.
For more information, see Rotate the CMEK for a disk or standard snapshot and Change the CMEK for a disk or standard snapshot.
Dataform
Feature
You can connect Dataform repositories to third-party Git repositories using Developer Connect, removing the need for manual secrets management and enabling support for repositories in privately hosted networks. This feature is generally available (GA).
Datastream
Feature
You can now create a Datastream stream directly from the overview page of your AlloyDB for PostgreSQL instance using the automated flow. The automated flow simplifies the process of moving data to BigQuery by reducing the number of steps that you need to perform.
For more information, see Create an AlloyDB for PostgreSQL stream using the automated flow.
Gemini Cloud Assist
Announcement
geminicloudassist API automatically enabled for Gemini Cloud Assist chat users
As of April 16th, 2026, the geminicloudassist.googleapis.com API has been
automatically enabled on projects that meet all of the following criteria:
- Had used Gemini Cloud Assist chat in the prior 60 days.
- Had the
cloudaicompanion.googleapis.comAPI enabled on April 16, 2026. - Did not have the
geminicloudassist.googleapis.comAPI enabled on April 16, 2026.
The Gemini Cloud Assist chat functionality that was previously served by
cloudaicompanion.googleapis.com is now served by
geminicloudassist.googleapis.com, and both APIs are dependencies to use
Gemini Cloud Assist. This automatic API enablement ensures that users have
access to the same functionality without any loss of service.
Memorystore for Valkey
Feature
You can migrate your workloads from your self-managed Redis and Valkey instances that run in Google Cloud Platform into Memorystore for Valkey. This feature is available in Preview.
Feature
The shared and customer-managed Certificate Authority (CA) modes are Generally Available.
Pub/Sub
Feature
You can stream messages from Pub/Sub directly to a Bigtable table using Bigtable subscriptions. This feature lets you write streaming messages to Bigtable without needing a separate subscriber such as Dataflow. This feature is available in Preview.
Security Command Center
Feature
AI Protection supports agentic workloads in Preview, including Vertex AI Agent Engine and Model Context Protocol (MCP) servers. This update includes the following:
- Agent vulnerability scanner: Identifies software vulnerabilities (CVEs) in workloads deployed with Agent Engine.
- Expanded detection and controls: Includes new threat detection findings and recommended security controls for AI agents and MCP servers.
- Enhanced inventory and filtering: Provides an updated AI security dashboard view and new filtering options for agentic resources in the console.
Source: Google Cloud Platform
