Delos Data raises $100 million to speed networking in AI data centers

Delos Data is targeting one of the ugliest bottlenecks in the race to AGI: shuttling data efficiently across increasingly heterogeneous clusters. As inference shifts from a handful of Nvidia GPU SKUs to a zoo of accelerators from Nvidia, AMD, Cerebras and others, the interconnect layer becomes the place where performance and energy efficiency are either unlocked or squandered. A startup that can abstract away that complexity with smart network silicon and control software immediately becomes strategically important to anyone trying to run large agentic workloads at scale.

From an AGI‑timeline perspective, this kind of infrastructure work quietly matters more than another flashy model demo. Scaling frontier systems will require clusters where thousands of accelerators stay fed without stalling on I/O or congested links. Delos Data is effectively offering a way to future‑proof AI data centers against uncertainty about the “right” compute mix, which encourages operators to keep investing aggressively in capacity instead of waiting for standards to settle.

It also underlines how much value is accruing in the AI plumbing around Nvidia, not just at Nvidia itself. If Delos can prove out its architecture, it strengthens the broader non‑Nvidia ecosystem and gives hyperscalers more flexibility to push down unit costs while still ramping total compute.