Research Notes

Google Cloud: 8th-Generation TPU Family Splits Training and Inference

Steve McDowell April 27, 2026

At Google Cloud Next 2026, Google unveiled its 8th-generation Tensor Processing Unit (TPU) family, splitting its custom silicon lineup into two purpose-built architectures for the first time.

The TPU 8t targets large-scale model training, emphasizing compute throughput and scale-up bandwidth, while the TPU 8i addresses inference and reasoning workloads, emphasizing memory bandwidth and low-latency communication.

Google Cloud Next ‘26: Storage Infrastructure Advances + the NetApp Partnership

Steve McDowell April 27, 2026

At its recent Next ’26 conference, Google Cloud advanced its storage portfolio across three fronts: high-performance infrastructure for AI training and inference, intelligent metadata capabilities built directly into the storage layer, and expanded ecosystem integrations.

The announcements span Cloud Storage Rapid, Google Cloud Managed Lustre with 10 TB/s throughput, Hyperdisk Exapools, Smart Storage automation, and material expansions to the Google Cloud NetApp Volumes service, including a deeper partnership with NetApp.

SUSECON 2026: SUSE Shows Off Platform Strategy and Agentic Infrastructure

Steve McDowell April 24, 2026

SUSECON 2026, recently held in Prague, saw SUSE lay out a coherent platform direction, shifting from being primarily a Linux and Kubernetes distribution vendor to a provider of an integrated open infrastructure stack spanning virtualization, edge, and AI-augmented operations.

CrowdStrike RSA 2026 Announcements: Securing Agentic Workflows

Steve McDowell April 16, 2026

CrowdStrike used the RSA Conference 2026 to announce a set of capabilities that directly address the problem of securing agentic workflows, based on the premise that the enterprise endpoint is the correct control plane for AI security governance.

MEXT: Predictive Memory Software to Control DRAM Costs

Steve McDowell April 10, 2026

MEXT, a Santa Clara-based startup founded just three years ago, recently launched Predictive Memory, a software-only solution that uses AI-driven prediction to extend effective server memory capacity by treating flash storage as a transparent extension of DRAM.

The company claims its product reduces infrastructure costs by 50% and increases usable memory capacity by 2x to 4x without requiring changes to hardware, operating systems, or applications.

Dynatrace Extends Observability Platform with Bindplane Acquisition

Steve McDowell April 8, 2026

Dynatrace has announced a definitive agreement to acquire Bindplane, an OpenTelemetry-native telemetry pipeline vendor formerly known as observIQ. Bindplane closes a gap in the Dynatrace platform at the data ingestion layer, extending its capabilities to edge-level telemetry collection, processing, and routing across hybrid and multi-cloud environments.