LDA Technologies launches 25G ultra-low-latency fabric for trading

LDA Maze delivers 3.6ns port-to-port latency across 160 links, targeting capital markets firms and HPC operators scaling to 25G.

By Disrupts Editorial Team Wed, 26 August 2026

A modern control room features a curved video wall displaying glowing blue data graphics across multiple screens, a matching curved conference table equipped with microphones and chairs, and bright natural light streaming from windows on th

LDA Technologies, the Mississauga-based ultra-low-latency networking specialist, has launched LDA Maze, a non-blocking full-mesh Layer 1 fabric it says delivers the highest link density available at 25G. The product offers 160 links of 10/25G connectivity, or 40 aggregated 100G links, with port-to-port latency measured between 3.6ns and 4.1ns. It is aimed primarily at capital markets trading architectures, but the company says the same fabric supports demanding high-performance computing and AI workflow applications.

Layer 1 fabrics sit below the IP networking stack, switching signals at the physical layer rather than routing packets. In latency-sensitive trading environments, where microseconds of advantage can translate directly into execution alpha, shaving nanoseconds at the interconnect layer is a meaningful engineering target. LDA positions Maze as infrastructure that allows firms to scale link density without increasing rack space, power draw, or cabling complexity.

Density without redesign

The product supports SFP, QSFP and QSFP-DD I/O cards in any combination, allowing the front-panel layout to match a firm's specific hardware build rather than the chassis design. Firms running 10G today can migrate to 25G on the same device without a hardware replacement, which LDA frames as a future-proofing argument against the capital expenditure of a full stack refresh.

Vahan Sardaryan, CEO and Co-founder of LDA, said: "There's been a gap in the market for high-density, 25G-capable Layer 1 technology, and LDA Maze addresses it. It provides a flexible, ultra-low latency interconnect between components within the chassis, front-panel links, FPGA boards, NICs and other equipment alike, so one fabric can power devices for finance, HPC and AI workflows."

LDA Maze is currently integrable with the company's existing Neo X platform. LDA says two further platform launches are expected in the coming weeks, at which point Maze will be available across the broader product line. The company has been operating since 2010 and positions itself as a specialist in FPGA-based networking for capital markets, with more than two decades of research and development behind its product set.

Convergence of trading infrastructure and AI compute

The release of Maze arrives at a moment when the infrastructure demands of electronic trading and those of AI workloads are converging more explicitly than at any previous point. Both disciplines require deterministic, nanosecond-range interconnects, dense link counts, and power-efficient rack designs. Historically, trading-grade low-latency hardware occupied a niche too specialised for general compute vendors to serve economically; AI's appetite for similar characteristics is beginning to reshape that dynamic.

For capital markets firms operating co-location environments at exchanges such as NYSE, Nasdaq, or the London Stock Exchange's Basildon data centre, the physical footprint of connectivity hardware is a direct cost denominator: rack space at co-location facilities is priced at a premium and is finite. A fabric that reduces cabling overhead while preserving latency performance addresses a genuine operational pressure, not merely a technical preference.

The broader investment landscape around trading infrastructure has remained relatively insulated from the wider technology venture cycle. Specialist providers like LDA, Solarflare (now part of Xilinx/AMD), and Arista's low-latency product lines compete in a market where switching costs are high and procurement decisions are driven by measured latency benchmarks rather than marketing claims. LDA's move to address a stated density gap at 25G suggests the firm is positioning ahead of the industry's gradual migration from 10G to 25G co-location architectures, a transition that parallel AI infrastructure build-out may now be accelerating. How quickly trading firms follow will depend on whether independent benchmarks validate the 3.6ns figure at scale across all 160 links simultaneously, a detail the release does not address.