Weekly Cloud Info #W11 - 2025

Author

Walid LARABI
March 15, 2025

Hi!

This week brings exciting updates from AWS, like availability of DeepSeek-R1 as a new fully managed AI model in Amazon Bedrock. GCP introduces tiered storage for Spanner to cut up to 80% of costs, while Azure expands event-driven capabilities in MySQL… and more.

Have a great read. ☕

📰 Top picks of the week

AWS is first cloud service provider to offer DeepSeek-R1 as fully managed, generally available model

AWS has launched DeepSeek-R1, a fully managed and serverless large language model (LLM) available in Amazon Bedrock. This model offers advanced reasoning and contextual understanding at a significantly lower cost compared to similar models. Customers can use DeepSeek-R1 for various complex tasks with built-in enterprise-grade security and the option to customize their models. Since its release, thousands of customers have adopted DeepSeek-R1 through the Amazon Bedrock platform.

AWS Introduces CodeBuild and Commands Rules for Enhanced CodePipeline Flexibility

AWS CodePipeline has added two new rules: CodeBuild and Commands, allowing customers to conditionally gate pipeline executions at the stage level. Users can start CodeBuild builds or execute shell commands based on the success or failure of actions in the stage. These features enhance flexibility and release safety by enabling integration tests and custom scripts during deployment. The new rules are available in all regions supported by AWS CodePipeline.

AWS AppSync Events Introduces WebSocket Publishing for Real-Time Applications

AWS AppSync Events now supports WebSocket publishing for real-time applications, allowing developers to create secure and efficient WebSocket APIs. This feature enables bi-directional communication over a single connection, reducing complexity and latency in apps like chat systems and multiplayer games. The enhancement is available in all supported AWS Regions. Developers can use any WebSocket client for implementation.

AWS Introduces ECR to ECR Pull Through Cache for Cost-Effective Image Syncing

Amazon ECR has launched a new feature called ECR to ECR pull through cache, allowing customers to sync container images between ECR registries across AWS regions or accounts. This feature reduces latency by caching only the images that are pulled, optimizing storage costs. Customers can configure rules to automatically fetch and cache images in downstream registries, ensuring they stay updated. The feature is available in all AWS regions, except GovCloud and China.

AWS Step Functions Enhances Local Development with Workflow Studio in VS Code

AWS Step Functions now offers an enhanced local IDE experience through the Workflow Studio integration in Visual Studio Code. This allows developers to build and edit state machines visually, mirroring the AWS Console functionality. With the AWS Toolkit extension, both JSON and YAML formats are supported, simplifying workflow development. The design mode features a graphical interface while the code mode allows direct editing in Amazon States Language.

AWS Enhances Amazon Cognito with Custom Access Tokens for Machine-to-Machine Authorization

Amazon Cognito now allows customization of access tokens for machine-to-machine (M2M) authorization flows. This enables fine-grained control over access for automated processes like data synchronization and microservices communication. Customers can add custom claims and scopes to access tokens, simplifying authorization logic in applications. This feature is available to Amazon Cognito customers using Essentials or Plus tiers in most AWS Regions, excluding GovCloud.

Azure Database for MySQL Introduces Event-Driven Triggers for Functions in Public Preview

Azure Database for MySQL now has triggers for Azure Functions in public preview, allowing developers to build event-driven applications. The triggers monitor MySQL table changes and enable functions to run when rows are created, updated, or deleted. This feature is only available for dedicated and premium Azure Functions plans. Change tracking requires altering table structures to include a timestamp column.

Azure Monitor Enhances Prometheus with HPA, Auto-Scaling 2 to 12 Replicas

Azure Monitor Managed service for Prometheus now features Horizontal Pod Autoscaling (HPA) for the ama-metrics pod. This allows automatic scaling based on memory usage, with a default range of 2 to 12 replicas. Customers can adjust the minimum and maximum replicas within this range. HPA helps prevent out-of-memory (OOM) kills.

GCP: Launch of Gemma 3 Model on Vertex AI with $300 Free Credits for New Users

Google Cloud has launched the Gemma 3 model on Vertex AI Model Garden, enabling easy fine-tuning and deployment. Users can leverage Parameter-Efficient Fine-Tuning (PEFT) and optimized vLLM deployment for enhanced performance with reduced resources. New customers receive $300 in free credits to test Gemma 3. The fine-tuning process requires a custom JSONL dataset uploaded to Google Cloud Storage.

GCP: Announcing Tiered Storage for Spanner - Cut Costs by 80%!

Google Cloud has introduced tiered storage for Spanner, enabling users to manage larger datasets efficiently. This new feature offers up to 80% cost savings by moving less frequently accessed "cold" data to cheaper HDD storage while maintaining a seamless experience. Automated processes simplify data management and ensure consistent query capabilities across storage tiers. Users can apply flexible policies to control data movement based on their needs.

❤️ You might also like

  • OpenStack comes to the Linux Foundation LINK

  • Gemini can now personalize its answers based on your search history LINK

  • First operating system for quantum networks unveiled LINK

  • Alibaba releases new AI model that can read human emotions LINK

  • OpenAI launches new tools for building AI agents LINK

  • Google quietly acquired 14% of Anthropic LINK

🏁 Enjoy this newsletter?

Forward it to a friend, and let them know they can subscribe here.