Google Cloud made U4 and its Ultra Low Latency Solution generally available, combining dedicated bare-metal compute with hardware-level multicast networking.

Both the U4 machine family and Ultra Low Latency (ULL) Solution reached general availability. Google Cloud says the solution, previously announced at Cloud Next ’26, targets high-frequency trading workflows, financial exchanges, and market participants. Its three stated infrastructure pillars are hardware-level networking, advanced network observability, and deterministic high-performance compute.

The networking design separates market feeds from management traffic. Two independent, dedicated Titanium adapters carry redundant A/B multicast market-data feeds, while a third adapter offloads telemetry and management and provides access to Google Cloud services. Hardware-level multicast ingestion bypasses traditional hypervisor-level virtual switches, and support for OpenOnload and the Data Plane Development Kit (DPDK) enables Linux user-space packet processing.

Compute and storage are specified at the host and persistent-storage levels. Dedicated bare-metal compute provides direct access to physical host resources, which Google says minimizes jitter for market-data processing and order routing. Local Titanium solid-state drives (SSDs) support transaction and tick logging, while Google Cloud Hyperdisk supports persistent market-data archives and analytics.

Observability includes hardware-accurate traffic capture and timing. Google describes dynamic capture for auditing, market replay, and trade validation without affecting primary traffic performance. Integration with its Firefly clock-synchronization system is stated to achieve sub-10-nanosecond network-level timestamping; this is a timestamping figure, not an end-to-end trade-execution latency guarantee.