Delos Data, a chip startup founded by Intel veterans, raises $100 million to fix the AI data center bottleneck GPUs can’t solve

Delos Data, a chip startup founded by Intel veterans, has raised more than $100 million to attack a problem threatening the economics of the AI boom: expensive GPUs and accelerators sitting idle waiting for data.

The startup is betting that the next big constraint inside AI data centers will come from the connections between chips rather than the chips themselves. That problem is becoming harder as data centers move beyond racks dominated by Nvidia GPUs and begin mixing hardware from Nvidia, AMD, Cerebras and other suppliers.

Matrix Partners, Playground Global, Socratic Partners, Capricorn’s Technology Impact Fund, Matter Venture Partners, IAG and DYNAMIQ participated in the funding round. Former Intel CEO Pat Gelsinger, now a general partner at Playground Global, is among the investors backing the company.

Delos Data plans to use the fresh capital to grow its hardware and software engineering teams and speed up product development and sales.

The pitch is straightforward. Buying more AI chips delivers little benefit if those processors spend too much time waiting for information to arrive.

“I can have all the computers in the world, and if I can’t get these devices communicating effectively with one another, I’ve got lots ⁠of hot hardware, burning watts and sitting here twiddling their thumbs,” Gelsinger said, according to Reuters.

AI’s bottleneck is moving between the chips

The first stage of the generative AI boom centered heavily on training huge models using clusters of Nvidia GPUs connected through Nvidia networking technology. AI infrastructure is becoming less uniform as companies deploy models at scale and AI agents handle longer, more persistent tasks.

Those workloads can span GPUs, CPUs, custom accelerators, memory and storage. Delos Data sees that growing mix as an opening.

“The most expensive idle asset in a data center is a GPU, CPU, or an accelerator waiting on the network. Inference workloads move data in a way that today’s interconnect was never designed to serve,” said Ed Doe, CEO and co-founder of Delos Data. “This funding positions the company to expand our team and offerings to solve the data transfer problem holistically by optimizing the network as the core of the AI system, thereby maximizing infrastructure ROI and ensuring that AI can scale to keep pace with massive growth in AI Inference.”

The company has developed networking chips and software to move data between those components with lower latency and greater efficiency, without locking customers into one type of compute hardware.

“We don’t know what the next infrastructure architecture is going to be for agentic (AI),” Delos Data Chief Technology Officer and Co-Founder Dan Daly said in an interview. “But we do know that we can provide the quickest, fastest way to move that data around.”

Delos Data on Tuesday introduced its Nonstop AI Data Interface, which the company says can deliver 10 times lower latency and 10 times higher efficiency. It joins the startup’s previously announced Nonstop AI Cluster and Nonstop AI Server products.

Those are company performance claims and will matter most once customers test the technology under large production workloads. Still, the problem Delos is targeting is real: AI infrastructure operators need to keep costly processors fed with data often enough to justify the enormous capital they’ve committed.

The economics get more pressing as AI shifts toward inference, where trained models respond to users and software agents perform tasks. Delos cites industry projections pointing to roughly 300 times more token demand by 2030, with agentic inference accounting for much of that increase.

A bet on a more fragmented AI data center

Delos Data’s opportunity rests partly on another shift taking place across the AI industry. Nvidia remains the dominant supplier of AI accelerators, yet customers have more choices than they did at the start of the generative AI boom. AMD has pushed deeper into AI infrastructure, Cerebras has gained attention for its wafer-scale systems, and hyperscalers continue developing custom silicon.

Nvidia itself now sells several types of processors for AI systems.

That creates a networking challenge. A data center filled with different processors needs a way to move huge amounts of information between them without leaving expensive hardware waiting.

“The most expensive idle asset in a data center is a GPU, CPU, or an accelerator waiting on the network. Inference workloads move data in a way that today’s interconnect was never designed to serve,” Delos Data CEO and co-founder Ed Doe said.

Delos brings some history to that bet. Its founders have spent years building data center silicon and systems, including work at Intel. Matter Venture Partners founding managing partner Wen Hsieh said the team has worked together on data center technology for two decades.

The $100 million round gives Delos substantial capital to prove that experience can translate into a new AI infrastructure business.

For years, the AI race was framed around who could build or buy the fastest processors. Delos is betting the next race will be decided by how quickly all those processors can talk to one another.

If that thesis holds, the industry may discover that owning more GPUs was only half the problem. Keeping them busy could be the harder half.