Delos Data Raises $100M to Fix AI's Networking Bottleneck

Delos Data, a Palo Alto-based startup, raised more than $100 million to launch Delos Nonstop AI, a full-stack approach to AI infrastructure built on the thesis that the network connecting GPUs together, not the chips themselves, is emerging as the true constraint on scaling AI inference, according to the company's own announcement of the round. Investors include Matrix, Playground, Socratic Partners, Capricorn's Technology Impact Fund, Matter Venture Partners, and IAG, alongside individual investors from the compute, hyperscale, networking silicon, and optical-connectivity industries.

Delos CEO and co-founder Ed Doe framed the underlying issue directly: "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." CTO and co-founder Dan Daly elaborated on why this problem is specifically getting worse as AI systems evolve: "Agentic AI is scaling beyond a single device, and when that happens performance isn't solely determined by the processors, it's decided by how the data moves between them." According to Delos's own framing, agentic AI workloads are fundamentally different from earlier AI usage patterns, since they run persistently across a growing mixture of GPUs, XPUs, and accelerators rather than completing in a single, discrete request-and-done cycle.

The company's product lineup spans multiple layers of infrastructure. Delos previously launched the Nonstop AI Server, a computing appliance optimized specifically for inference that customers can equip with up to four graphics cards from external suppliers, according to SiliconANGLE's reporting on the raise. This new funding expands that lineup with the Nonstop AI Data Interface and a Nonstop AI Reference Architecture, offered across three formats: a PCIe card for existing servers, a chiplet capable of processing more than 30 terabits of traffic per second for direct integration into AI accelerators, and a near-packaged optics version offering more than 10 terabits per second designed to sit on the same substrate as the accelerator using it.

Delos Data's Nonstop AI Portfolio at a Glance

Socratic Partners general partner and co-founder Christian Gallagher framed the investment thesis around a shift in where AI infrastructure value is actually concentrating. "Compute has been the story for over a decade, but the return on an inference cluster is now set by what happens between the accelerators, and how much capital sits idle as insurance against failure," Gallagher said, according to Delos's own announcement. That framing, treating idle, underutilized compute capacity as a genuine, quantifiable financial cost, connects directly to the broader AI infrastructure efficiency investment pattern we've tracked closely this month, including Lumilens' own $900 million raise targeting the exact same GPU connectivity bottleneck, and Xsight Labs' $300 million raise for AI networking chips.

This raise lands directly alongside independent market data confirming the underlying trend Delos is betting on. Gartner forecasts worldwide AI spending will total $2.7 trillion in 2026, a 49.5% increase year over year, with demand for AI infrastructure, including AI-optimized servers and network fabric, described as "strong and inelastic to pressures from memory-related pricing increases," according to Gartner distinguished VP analyst John-David Lovelock. Separately, McKinsey projects inference could account for more than 40% of total data center demand by 2030, according to TechEdgeAI's reporting, reinforcing the specific bet Delos and its investors are making about where genuine infrastructure investment needs to concentrate going forward.

Not every account of this raise treats the company's technical claims as fully settled fact. TechEdgeAI's own reporting was direct in noting Delos's claimed 10x improvements in latency and efficiency remain "subject to independent benchmarking," a genuinely useful reminder that vendor-reported performance figures, however impressive, still require third-party validation before being treated as confirmed industry benchmarks.

Delos Data's raise is worth understanding for any business evaluating AI infrastructure vendors or planning large-scale AI deployment specifically, since it reinforces a genuine, increasingly well-documented shift in where the real bottleneck in AI scaling actually sits, moving from pure chip supply toward the networking and data movement layer connecting those chips together at scale.

For businesses planning long-term AI infrastructure investment, the specific detail that idle, network-bottlenecked compute capacity represents genuine wasted capital is worth factoring directly into procurement and deployment planning, since this represents a real, quantifiable cost separate from the sticker price of the underlying chips themselves.

What does Delos Data's Nonstop AI technology actually do?
Delos Data builds infrastructure specifically designed to solve data movement bottlenecks between AI chips, offering a portfolio spanning servers, a data interface chiplet, near-packaged optics, and reference architecture, all aimed at keeping AI accelerators fully utilized rather than idle while waiting on network transfers.

How much funding has Delos Data raised?
Delos Data raised more than $100 million from investors including Matrix, Playground, Socratic Partners, Capricorn's Technology Impact Fund, Matter Venture Partners, and IAG, alongside individual investors from the compute and networking industries.

Why does Delos Data say networking is becoming AI's next major bottleneck?
The company argues that as agentic AI workloads run persistently across a growing mixture of GPUs and other accelerators, performance increasingly depends on how efficiently data moves between those chips, rather than purely on chip processing power itself, with idle accelerators waiting on network transfers representing genuine, wasted infrastructure capital.

Delos Data raised more than $100 million to launch Delos Nonstop AI, a full-stack infrastructure portfolio betting that the network connecting AI chips together, not the chips themselves, represents the industry's next major scaling bottleneck. The company's technology spans a PCIe card, a chiplet processing more than 30 terabits per second, and near-packaged optics offering more than 10 terabits per second, with claimed 10x improvements in latency and efficiency that remain subject to independent benchmarking. The raise lands alongside Gartner's forecast of $2.7 trillion in worldwide AI spending for 2026 and McKinsey's projection that inference could represent more than 40% of total data center demand by 2030, reinforcing the broader industry shift toward prioritizing infrastructure efficiency over raw chip supply alone.