60 Microseconds to Market: How Databento Achieves Financial Industry-Leading Latency

Databento adopted a comprehensive NVIDIA-powered networking stack to address its critical business and technical challenges. Leveraging NVIDIA Spectrum™ Ethernet switches and ConnectX SuperNICs across its private cloud infrastructure spanning four data centers with over 30 PB of storage capacity, Databento built a high-performance, cost-efficient, and scalable network foundation.

A pivotal change was implementing NVIDIA Cumulus Linux for network switch management, letting Databento automate and control its network with the same tools as its servers. This simplification reduced operational complexity and cut network operations overhead by over 60%, helping alleviate the workload on specialized networking staff.

For real-time market data, NVIDIA ConnectX NICs paired with technologies like VMA and DPDK delivered ultra-low latency packet processing—helping Databento outperform traditional and expensive proprietary networking gear while maintaining deterministic, jitter-minimized forwarding from host to host using NVIDIA Spectrum switches and ConnectX NICs.

Further scaling was enabled by software-defined networking through the Netris overlay on NVIDIA hardware, creating a private cloud network environment that elastically grows with business demands. Netris software gateways transformed commodity servers equipped with NVIDIA NICs into multifunctional routing, firewall, and load-balancing devices, replacing costly, fixed-function appliances and significantly reducing infrastructure costs while keeping the network consistent, predictable, and intent-driven as Databento grows.

With NVIDIA's advanced networking solutions, Databento achieved extraordinary results that powered its explosive growth and innovation in financial market data services:

• Accelerated data delivery: Databento's median network latency dropped below 60 microseconds using NVIDIA NICs with DPDK acceleration. This unparalleled speed delivered U.S. options market data to customers within 242 microseconds over the internet-a game-changing capability that lets customers eliminate expensive physical colocation and cut hosting costs by over 80%, accelerating migration to cloud-based architectures. NVIDIA Spectrum silicon and ConnectX NICs provide deterministic low latency with consistent, jitter-minimized forwarding between hosts, even during extreme market volatility.
• Zero- or low-loss market-data delivery: Spectrum's advanced buffer management and NVIDIA Cumulus telemetry help Databento maintain zero or near-zero packet loss during intense market spikes, protecting downstream trading strategies from missing critical ticks while sustaining very high throughput.
• Precision timekeeping and ordering: ConnectX hardware timestamping enables highly accurate sequencing of market-data events, improving both internal processing pipelines and the ordering guarantees Databento provides to customers consuming its feeds.
• Reduced CPU overhead and higher throughput: Hardware offloads and kernel-bypass paths on NVIDIA NICs allow Databento to maximize packet processing performance while minimizing CPU load, freeing cores for analytics, AI, and customer-facing workloads.
• Deep observability and operational efficiency: Telemetry from NVIDIA Spectrum switches, surfaced through tools such as Prometheus and Grafana, gives Databento fine-grained visibility into microbursts, asymmetric traffic flows, and transient congestion, enabling rapid diagnosis and resolution of anomalies.
• Streamlined operations and scalability: Network automation and unified management with Cumulus Linux and Netris slashed Databento's operations overhead by more than 60%, enabling a lean team to keep pace with a 3,800% annualized revenue run rate growth over two years, scaling infrastructure seamlessly alongside business and headcount expansion. Cumulus Linux on NVIDIA Spectrum gives Databento an open, programmable fabric with deep, hardware-level visibility for ultra-low-latency trading.
• Optimized cost and performance: Using NVIDIA Ethernet switches and NICs, Databento optimized its 100G NVMe storage network for both performance and cost-efficiency, achieving similar levels of performance at roughly 20% lower cost versus other vendors.

These improvements strengthened Databento's competitive edge in serving top-tier clients across financial markets, hedge funds, AI companies, and fintech firms, cementing its position as an industry disruptor.