Delos Data raises $100M to keep AI data center chips from idling
What's the deal? Delos DataDealroom has a profile for this one. Try Dealroom →, a chip startup founded by Intel veterans, raised $100 million to build networking chips and software that help AI data centers avoid leaving expensive processors idle while they wait for data.
What's the endgame? The company is designing networking chips and software meant to keep mixed fleets of GPUs and other accelerators talking to each other faster. The goal: stop costly processors from sitting idle as they wait for data.
Why now? Early AI data centers ran mostly Nvidia graphics processing units (GPUs) tied together with Nvidia's own networking gear, ReutersDealroom has a profile for this one. Try Dealroom → reported. That kept things simple when the main job was training big models.
What's changing? Operators are now adding different types of accelerators — specialized chips used to run AI faster — from rivals such as Advanced Micro DevicesDealroom has a profile for this one. Try Dealroom → and Cerebras Systems, plus multiple chip lines from Nvidia. Mixing hardware makes moving data between chips harder.
The signal: As AI workloads shift beyond pure training, the bottleneck is moving from raw compute to how efficiently chips connect. Delos Data is betting that vendor-neutral networking, not any single chipmaker's stack, is where the next wave of data center spending flows.
Image credit: Delos Data
