Is Speed Still the Point of a Low-Latency Trading Stack?

For two decades, capital markets technology was defined by the goal of building the fastest low-latency stack possible. Today, that thinking is no longer enough. As our CTO, Deepak Dhayatker, highlighted at A-Team Group’s ExchangeTech Summit London in June, firms that still view competitiveness purely through the lens of speed risk falling behind.

Speaking on the panel “Building the Ultimate Low Latency Highway: Execution Gateways, Order Routing and Ensuring Near Zero Jitter and Determinism,” Deepak joined Anvar Karimson (Kepler Cheuvreux), Irina Sonich-Bright (UBS), and Tom Swarbrick (Exegy) to challenge some long-held assumptions about market microstructure technology.

Speed Is the New Baseline

An audience poll asked where low latency still contributes most to alpha generation. Order execution and routing ranked ahead of market data processing and signal generation.

The discussion quickly revealed that the answer depends on the business model. A proprietary trading firm running FPGA infrastructure has very different priorities from an agency broker supporting multiple trading desks. For many brokers, competitive advantage increasingly comes from capabilities that extend beyond raw speed.

The panel agreed that tick-to-trade latency is only as strong as its weakest link. Improving one part of the chain while ignoring others delivers limited benefits. As Irina Sonich-Bright noted, executing quickly on poor-quality data offers little value.

Determinism Matters More than Raw Speed

Software provides flexibility but is inherently less deterministic. FPGA hardware offers greater consistency, but its cost, complexity, and specialist skill requirements put it beyond the reach of many firms.

Irina compared the challenge to driving a Ferrari in heavy traffic. For benchmark-driven strategies such as VWAP, understanding the route and anticipating delays matters more than accelerating at every opportunity. Paying for FPGA infrastructure where microseconds do not affect outcomes rarely makes commercial sense.

Agility Has Become a Competitive Advantage

Anvar Karimson argued that the ability to change a system is now often more valuable than the ability to run it faster. Market structure continues to evolve through bilateral trading, systematic internalisers, extended trading hours, and new venues and order types. Agency clients increasingly expect brokers to support these changes quickly.

That expectation creates tension with highly specialised hardware, which can be slower to adapt. A system that is marginally slower but easier to evolve may provide more long-term value than the fastest infrastructure available.

Low Latency Is a Discipline, Not a Component Choice

Deepak’s key message was that low latency is not a product you buy, but an engineering discipline that requires optimisation across the entire stack.

The guiding principle is the Theory of Constraints: every system has a small number of bottlenecks, and improving anything else delivers limited gains. Effective optimisation starts with profiling, measurement, and the systematic removal of the largest bottleneck.

He highlighted several factors that contribute to performance, including colocation, proximity cloud services, precision time protocols, thread pinning, interrupt isolation, and careful NUMA configuration. Emerging technologies such as CXL 3 remote memory could also reduce latency jitter by minimising garbage collection issues.

At the software layer, tools such as DPDK, Solarflare Onload, zero-copy architectures, and non-blocking data structures all play a role. Just as importantly, observability must be built into the platform from the start. Without clear visibility into the pipeline, meaningful optimisation is impossible.

FPGA: Use It Where It Counts

For high-frequency proprietary trading, FPGA remains the fastest and most deterministic option for market data processing and execution. However, the panel repeatedly returned to the same challenges: cost, specialist skills, and reduced flexibility. Deepak argued that firms do not need FPGA everywhere. The greatest value comes from deploying it at the highest-throughput bottlenecks and using software for enrichment, normalisation, and other less latency-sensitive functions. Well-optimised Java and C++ applications can handle much of this workload effectively.

Tom Swarbrick noted that this approach is influencing product development across the industry. Hardware acceleration is increasingly being packaged for broader agency use cases, with configuration exposed through higher-level tools rather than specialist hardware languages.

Regulation Is Changing the Equation

One panellist argued that MiFID II has reduced the venue competition created by MiFID I, leaving firms often trading at the speed of the slowest relevant market. The continued success of exchanges using speed bumps suggests that engineered fairness does not necessarily reduce market participation.

DORA also featured prominently. Anvar highlighted that firms remain responsible for the resilience of critical third-party providers, making vendor selection a key consideration for EU-regulated organisations. At the same time, AI regulation, MiCA, and digital asset frameworks are adding further complexity that firms must be able to absorb quickly.

Where the Edge Moves Next

As market access becomes increasingly standardised, firms have fewer opportunities to differentiate through infrastructure alone. Deterministic connectivity, low-latency networks, and proximity hosting are no longer exclusive advantages. The panel pointed to three areas where firms can still create meaningful differentiation:

• Agility in adapting to new venues and order types
• Cost discipline across infrastructure and execution
• Intelligence through analytics, algorithms, and decision-making

At Rapid Addition, this thinking shapes how we build execution gateways and order-routing technology. Performance still matters, but speed alone is no longer enough. Firms need infrastructure that can adapt quickly, operate efficiently, and support increasingly sophisticated execution strategies. Low latency remains essential, but it is now one part of a much broader competitive equation.