Delos Data's $100M Gambit to Build Unbreakable AI Infrastructure

PALO ALTO, CA – September 15, 2026 – In a significant move to reshape the foundational layers of artificial intelligence, Delos Data today announced it has secured over $100 million in new funding and unveiled a comprehensive portfolio, Nonstop AI™, aimed at solving a critical and growing bottleneck in AI performance: the network itself. The company, founded by Intel veterans, is targeting the complex infrastructure needs of next-generation “agentic AI” with a new reference architecture and data interface designed to deliver truly continuous, resilient, and scalable AI operations.

The funding round, which brings Delos Data’s total capital raised to over $130 million, saw participation from a formidable syndicate including Matrix, Playground, Socratic Partners, and Capricorn's Technology Impact Fund, among others. The investment underscores a growing conviction within the venture and technology communities that as AI models become more sophisticated, the infrastructure supporting them requires a fundamental redesign.

“There has been real excitement around optical links, new standards, and die-to-die interconnects, each one solving a real piece of the puzzle,” said Ed Doe, CEO and co-founder of Delos Data, in a statement. “What's missing is the network in aggregate, rethought for a workload that didn't exist when these networks were designed. Agentic AI is always on, always running, always being fed. You can't get there by applying the old network to a new problem.”

The Agentic AI Challenge and the 'MoXI' Problem

Delos Data’s strategy is a direct response to the operational shift from request-and-response AI models to persistent, autonomous agentic AI systems. Unlike their predecessors, these agents are designed for continuous operation, performing complex, multi-step tasks over extended periods. This “always-on” nature creates unprecedented demands for infrastructure that is not only fast but exceptionally resilient. Industry analysts like Gartner have projected that a significant number of organizations using agentic AI will face operational incidents by 2029 due to inadequate runtime controls, highlighting the urgent need for robust infrastructure.

The core challenge, as Delos Data frames it, is the “Mixture of X Infrastructure” (MoXI). A single agentic AI workload runs across a heterogeneous mix of hardware (GPUs, CPUs, accelerators), various AI models, and a complex web of switches, links, and network topologies. This fragmented reality, born from retrofitting data centers designed for CPU-centric tasks, creates what the company calls the “Interconnect Wall.” AI accelerators, despite their power, often sit idle, waiting for data to traverse this inefficient and brittle network, leading to underutilization and soaring operational costs. The focus is shifting from raw compute power to the economics of token generation, or “Tokenomics,” where the cost per token is the ultimate metric of efficiency.

A New Blueprint for Nonstop Tokens

To dismantle this wall, Delos Data has introduced a system-level blueprint, the Nonstop AI™ Reference Architecture, and its linchpin component, the Nonstop AI™ Data Interface. The architecture’s central innovation is the introduction of a universal data interface between every endpoint—be it a GPU, memory, or storage device—creating a unified data domain. This allows for the flexible composition of diverse hardware resources tailored specifically to the workload, with the goal of delivering what the company calls “Nonstop Tokens.”

The Data Interface itself is a key piece of operational innovation, delivered in three distinct form factors to serve every class of endpoint: a high-speed I/O chiplet for accelerators, near-packaged optics integrating optical engines, and a PCIe card for CPUs and memory. This interface sits between the endpoint and the network, acting as the first line of defense against failures.

“These engines spend most of their time waiting to be fed,” noted Dan Daly, CTO and co-founder of Delos Data. “When you're stitching together endpoints from different vendors at this scale, you can't engineer failure away, you must build for it from the beginning.”

This is where Delos Data makes its boldest claims. The company targets a 10x improvement in performance (latency), a 10x increase in scale (supporting up to 100,000+ endpoints versus the dozens in current high-end systems), and a 10x boost in resiliency. The resilience is not just a software promise; it’s baked into the Delos Apollo Data Interface silicon, which manages load balancing, topology mapping, and failure recovery directly in hardware. Whether a link breaks, an accelerator dies, or a software update causes a hiccup, the system is designed to reroute traffic and continue operating without stopping the workload—a capability being demonstrated live this week at the AI Infra Summit.

The $100 Million Bet on Holistic Design

The substantial investment in Delos Data is a calculated bet that a holistic, system-level approach is the only way to solve the AI networking crisis. Investors are backing a team with deep experience in data center silicon and systems to tackle a problem that piecemeal solutions have failed to address. The participation of industry luminaries alongside top-tier VCs signals a shared belief that the company’s integrated hardware and software strategy can unlock significant ROI for the builders of next-generation AI.

This strategy places Delos Data in a competitive landscape that includes established interconnects like InfiniBand and proprietary solutions like NVIDIA's NVLink, which is currently limited in its scale-up capacity. By designing a protocol-agnostic solution that supports a wide array of physical transport options—from copper to co-packaged optics and emerging standards like Ultra Ethernet (UEC)—the company is positioning itself not as a replacement for one specific technology, but as a unifying fabric for the entire AI cluster. This approach could appeal to hyperscalers, AI-focused cloud providers, and large enterprises looking to escape vendor lock-in and build more flexible, future-proof infrastructure in a market for AI interconnects projected to surpass $100 billion by 2030.

From Co-Design to Deployment

Delos Data is moving aggressively from concept to commercial reality. Its Delos Morpheus Platform, a hardware and software development kit, is available now, allowing customers to co-design and simulate their cluster topologies before committing to a physical build. This enables infrastructure teams and model builders to work together, optimizing the system for specific workloads and failure modes in a virtual environment.

The Delos Nonstop AI™ Clusters, which can run on existing infrastructure, are already in production with select customers, as is the Delos Mosaic software. The next major milestone will be the sampling of the Delos Nonstop AI™ Server, an Open Rack v3 compliant design, to customers by the end of 2026. By demonstrating live failure recovery at major industry events and engaging customers in a co-design process, Delos Data is methodically building the case that its vision for a faster, stronger, and more efficient AI infrastructure is not just a blueprint, but an executable plan.