Hyperscalers, challenger clouds and the economics of running compute at scale.
Rising memory costs are pushing Cloudflare to rethink how it stores data across its distributed CDN. A new prototype trades modest CPU overhead for petabytes of recovered storage capacity by compressing eligible assets with Zstandard before they hit disk.
Anthropic is maintaining Claude access for Cursor despite SpaceX's acquisition, a stark reversal from its treatment of Windsurf and xAI—a decision observers attribute to Anthropic's reliance on SpaceX's Colossus data center.
Amazon's new R9g and R9gd EC2 instances, built on Graviton5 processors, are now available across multiple regions. The instances deliver significant performance improvements over their Graviton4 predecessors while maintaining the energy efficiency AWS claims for its custom silicon.
Amazon's flagship compute service marks two decades since its 2006 launch with a portfolio spanning custom silicon for AI training and inference, Mac instances for Apple developers, and deployment models extending from 39 global regions into edge and 5G networks.
AWS introduced student rewards on its Builder Center, opened a new Local Zone in Las Vegas, and added batch instance termination to EC2 Auto Scaling. The week also brought updates to CloudShell, Bedrock model support, and AWS Glue 6.0 with significant pricing improvements.
AWS rolled out application-level status checks for EC2 instances, an automated IAM role manager, and access to OpenAI's Daybreak cybersecurity tools on Amazon Bedrock, alongside new foundation models in SageMaker JumpStart.
The isp6 team deployed their order-management platform as an IPv6-first Kubernetes cluster on AWS and documented which dependencies still force IPv4 back into the architecture, what they learned about subnet sizing, and whether the financial case holds up in 2026.
Cloud infrastructure provider Nscale has secured a definitive agreement to buy Anyscale, the AI workload scaling platform, raising questions about whether cloud-agnostic AI software can maintain vendor neutrality when bundled with GPU resources.
