Delos Data secures venture funding to develop resilient, low-latency interconnects tailored for agentic AI workloads.
Delos Data, a Silicon Valley start-up founded by former Intel engineers, has raised over $100 million in funding to develop next-generation networking chips for data centres. The investment round was backed by Matrix, Playground, Socratic Partners, Capricorn’s Technology Impact Fund, Matter Venture Partners, and IAG.
The capital will be used to expand the company’s hardware and software engineering teams, accelerate product development, and scale commercial sales.
Delos Data argues that the rise of “agentic” AI—where models run continuously across interconnected clusters rather than answering simple, isolated queries—will overwhelm existing data centre networks. Industry projections suggest token demand could grow 300-fold by 2030, with persistent agentic inference driving the vast majority of this traffic.
The firm is pioneering an architecture dubbed “Nonstop AI”, operating on the belief that networking—rather than raw compute power, memory, or electricity—presents the primary bottleneck for future AI infrastructure. Delos Data CEO and co-founder Ed Doe highlighted that GPUs and AI accelerators frequently sit idle whilst waiting for network transfers, as traditional interconnects were never designed for complex inference data paths.
To resolve this, Delos Data’s platform bridges diverse hardware components—such as GPUs, custom accelerators, CPUs, and flash storage—without requiring high-latency conversion steps. The firm claims its products operate with latency of around 100 nanoseconds, compared with several microseconds for standard network interface cards (NICs).
To help clients test the system, Delos Data has launched Morpheus, a PCIe development card that enables customers to validate their intellectual property against live workloads across various network topologies.
Beyond speed, CTO and co-founder Dan Daly emphasised that the platform integrates end-to-end hardware resilience directly into the data interface. This ensures reliable data transfer across massive clusters, preventing network failures without forcing processor designers to build complex error recovery directly into their chips.
Ultimately, Delos Data aims to free AI developers from hardware vendor lock-in, allowing organisations to combine best-in-class components into flexible, co-designed infrastructure topologies.
