The round
Delos Data, a Palo Alto start-up founded by former Intel engineers, said on Tuesday that it had raised more than $100 million to develop networking chips and software for AI data centres. In its announcement, the company named Matrix, Playground, Socratic Partners, Capricorn’s Technology Impact Fund, Matter Venture Partners and IAG as investors, alongside unnamed individual backers from across the industry. Reuters also listed DYNAMIQ among the investors. Neither the company nor Reuters gave a valuation or named the round.
Delos said the money would pay for more software and hardware engineers and speed up product development and sales. The company is run by its co-founders, chief executive Ed Doe and chief technology officer Dan Daly.
The pitch
Delos argues that the network, rather than power, memory or compute alone, is becoming the chokepoint in AI. As data centres shift from training models to serving AI agents, Reuters notes, they increasingly mix Nvidia GPUs with chips from rivals such as Cerebras and AMD. Delos wants to connect whatever mix of hardware an operator chooses.
“The most expensive idle asset in a data center is a GPU, CPU or an accelerator waiting on the network,” Doe said in the release.
Pat Gelsinger, Intel’s former chief executive and now a partner at Playground, made a similar point to Reuters. He described processors left waiting for data as one of the biggest wastes of money and energy in AI data centres.
“We don’t know what the next infrastructure architecture is going to be for agentic (AI),” Daly told Reuters, adding that Delos could offer the fastest way to move the data.
The products
Alongside the funding, Delos announced a Nonstop AI Data Interface and a reference architecture, adding to cluster and server products it showed earlier this year. The Data Interface comes in three forms: an I/O chiplet for GPUs and other accelerators, which Delos rates at more than 30 terabits per second; near-packaged optics; and a card for CPUs, flash storage and memory. Delos says the interface detects failures and handles recovery in hardware.
The company says it is targeting 10 times faster performance, 10 times stronger resiliency and 10 times higher scale than endpoints achieve today. In the funding release it also claimed 10 times lower latency and 10 times higher efficiency. These are Delos’s own figures, and no independent test results have been published. Doe told Electronics Weekly that latency would be around 100 nanoseconds, against microseconds for a typical network interface card.
Where it stands
According to Delos, its Nonstop AI Clusters platform is in production and available now for customers designing their next clusters. The Nonstop AI Server, shown at Computex earlier this year, is due to sample to customers at the end of 2026. The company did not give a date for the Data Interface silicon.
What to watch
The server samples at the end of the year are the next milestone Delos has named. Until customers test the hardware, its performance targets remain claims.
