At Google Cloud Next ’26, Google Cloud announced the general availability of its Ultra Low Latency (ULL) Solution for high-frequency trading, together with the new U4 machine family that powers it. The solution targets capital markets where operational success depends on predictable performance and low latency, and where even tiny variations in packet timing can alter queue priority and trade execution.

The motivating problem is twofold: trading venues and participants need deterministic, co-location-grade speed, but traditional on-premises data centers are hitting limits on power, physical rack space, and hardware procurement cycles. The ULL Solution aims to combine the speed and determinism of physical co-location with cloud scalability and automation. It is built on three infrastructure pillars: hardware-level networking for scalable, hardware-based multicast distribution of market-data feeds; advanced networking observability with hardware-level timing accuracy for auditing, market replays, and trade validation without impacting primary traffic; and deterministic high-performance compute for latency-critical execution tiers.

The resulting capabilities are broad. U4P and U4C are dedicated bare metal instances that provide direct access to physical host resources, minimizing jitter. Local Titanium SSDs handle real-time transaction and tick logging, while Google Cloud Hyperdisk provides persistent archives and analytics storage. Traffic isolation is physical: redundant A/B multicast feeds arrive through two independent Titanium adapters, while a third adapter handles telemetry, management, and access to Google Cloud services. Hardware-accelerated multicast ingestion bypasses hypervisor-level virtual switches, and support for OpenOnload and DPDK delivers predictable user-space packet processing with minimal application changes. Firefly clock synchronization provides sub-10 nanosecond network-level timestamping, comfortably better than the sub-100 microsecond UTC synchronization regulatory requirement for financial exchanges. Built-in telemetry offers 24/7 low-latency packet capture and out-of-band packet brokering for compliance and analytics. A 24-7 market-ready design isolates production workloads in a dedicated primary zone, with maintenance and qualification testing in a secondary zone. Trading teams can offload data streams over an independent network interface to Google Cloud services such as BigQuery and Gemini Enterprise.

The U4 family comes in three series. U4P and U4C are dual-socket bare metal instances built on Intel 5th Gen Xeon (Emerald Rapids) with 120 physical cores, 512 or 768 GB of memory, and three physical NICs (one standard, two ULL); U4P targets exchange operators and U4C targets market participants. The U4S VM series runs on Intel 6th Gen Xeon (Granite Rapids) with 2 to 288 vCPUs, up to 2,232 GB of memory, multi-vNIC networking up to 200 Gbps, and larger local Titanium SSD capacity, aimed at workloads like pre-trade risk validation and real-time analytics deployed alongside the bare metal tiers.

The post highlights an ongoing collaboration with CME Group to migrate listed derivatives markets to Google Cloud. CME’s Managing Director of Markets Engineering, Pearce Peck-Walden, says the U4 instances combine deterministic low-latency networking and dedicated bare metal compute with the benefits of the cloud, letting participants ingest market data and execute trades with co-location-like speed and cloud agility. Additional testing quotes come from Prime Trading’s EVP/CTO Corbin Kidd, who noted that having both bare metal and VM options helps align infrastructure cost with strategy, and 28Stone’s Christopher Wilson, whose tick-to-trade stress testing showed a virtually flat performance curve, which he attributes to structural determinism rather than raw speed.

Storage and placement are also covered: U4P and U4C offer up to 12 TiB of direct-attached local Titanium SSD, U4S scales to 18 TiB, and Hyperdisk Balanced and Extreme volumes scale to 32 volumes and 512 TiB per instance. Compact placement policies reduce physical network hops, while spread placement policies improve resiliency. The ULL Solution is available in select private Google Cloud regions, with access by request. Google positions the architecture as a future foundation for algorithmic trading, feeding real-time markets directly into AI-driven trading models.