Xelera Technologies continues to engage with the global STAC community, sharing insights and technology designed to bring machine learning inference closer to the execution path in performance-critical financial markets. Across events in London, Hong Kong, and New York, the team has explored how CPU optimization, FPGA acceleration, and flexible deployment architectures can help trading firms meet demanding latency and scalability requirements.
At STAC Summit London, Xelera joined industry discussions on high-performance trading technology, low-latency AI inference, and infrastructure for financial markets.
A key topic was the role of CPU-based inference in latency-sensitive environments. Intel’s Evgeny Khartchenko presented “AI with Intel® Xeon® 6 Processors,” covering neural network inference using Intel® Xeon® 6 processors and Advanced Matrix Extensions (AMX). The session included benchmark results from Xelera Technologies’ CPU-only neural network inference engine, demonstrating an approach to optimizing inference performance without requiring FPGA acceleration.
Represented by Andrea Suardi, Xelera used the event to exchange perspectives on CPU-based acceleration, AI deployment, and the evolving infrastructure requirements of modern trading systems.
At HKEX Connect Hall, Xelera participated in discussions on low-latency trading, data center capacity planning, and the changing needs of financial markets across the Asia-Pacific region.
A highlight was the presentation by Andrea Suardi, Head of Acceleration Technology at Xelera Technologies, titled Xelera Silva: Sub-Microsecond ML Inference — Three Execution Modes, One API.
The session explored how Xelera Silva enables trading firms to deploy neural networks and decision trees using FPGA acceleration, without requiring extensive specialist hardware expertise. With sub-microsecond p99 inference latency as a key performance target, the presentation highlighted the potential of bringing machine learning directly onto the execution path, where predictable performance and minimal latency are essential.
The event also provided an opportunity to discuss emerging market structures, including digital assets, and the infrastructure needed to support increasingly demanding trading workloads across APAC.
At STAC Summit New York, Xelera continued the conversation around applied AI, quantitative workflows, low-latency inference, and performance-focused trading infrastructure.
Andrea Suardi represented Xelera Technologies and presented Sub-Microsecond ML Inference — Three Execution Modes, One API, highlighting the flexibility of Xelera Silva for financial institutions seeking to integrate machine learning into latency-critical trading systems.
Xelera Silva supports three execution modes through a single API: CPU-Only, Accelerator mode, and Inline Silicon. This architecture allows teams to adapt inference deployment to different latency, concurrency, and infrastructure requirements while maintaining the same trained model and API interface.
The CPU-Only approach enables optimized inference on general-purpose processors, while accelerator-based execution provides additional options for demanding workloads. Inline Silicon offers another deployment path for applications requiring inference to be integrated more closely with the execution infrastructure.
Together, these options give trading teams greater flexibility in balancing performance, scalability, and deployment complexity as they move machine learning from research into production.
Across these STAC events, a common theme emerged: the value of machine learning in financial markets depends not only on model quality, but also on how efficiently and reliably those models can be executed in real-world trading environments.
From CPU-only neural network inference to FPGA-accelerated neural networks and decision trees, Xelera Technologies is focused on making high-performance AI deployment more practical for financial institutions.
Through Xelera Silva’s three execution modes and unified API, the company aims to help teams align inference technology with their specific performance requirements without unnecessarily complicating model deployment.
Participation in the STAC community provides a valuable opportunity to exchange technical insights with financial institutions, infrastructure specialists, and technology partners, contributing to the ongoing development of faster, more scalable, and more reliable AI-powered trading systems.
Xelera Technologies looks forward to continuing these conversations with the global financial technology community as demand for low-latency machine learning and performance-critical infrastructure continues to grow.