Delos Data Closes Over $100M to Optimize AI Data Flow

The big picture: Delos Data, a chip startup founded by Intel veterans, has raised over $100 million to address a critical bottleneck in AI data centers: expensive GPUs and accelerators sitting idle waiting for data.

Why it matters:

• Idle GPUs: Costly AI processors are frequently underutilized because they spend too much time waiting for data to arrive, severely impacting return on investment for AI infrastructure.
• Diverse Hardware: AI data centers are evolving beyond uniform Nvidia GPU racks, incorporating a mix of hardware from Nvidia, AMD, Cerebras, and custom silicon, which complicates efficient data transfer between components.
• Inference Shift: The industry’s pivot towards AI inference and agentic workloads demands significantly more efficient and lower-latency data movement, a challenge current interconnect technologies were not designed to meet.

How it works:

• Networking Chips & Software: Delos Data develops specialized networking chips and software designed to move data between diverse AI components with lower latency and greater efficiency.
• Nonstop AI Interface: The company introduced its Nonstop AI Data Interface, claiming to deliver 10 times lower latency and 10 times higher efficiency for data transfer within AI systems.
• Vendor Agnostic: Delos Data’s solution aims to optimize data flow across heterogeneous AI infrastructure, preventing vendor lock-in and maximizing the utilization of various compute hardware.

The catch: Delos Data’s performance claims, while promising, will require validation under large-scale production workloads by customers. The market for AI interconnects is increasingly competitive, with established players like Nvidia and emerging solutions from other chipmakers and hyperscalers also vying to solve data transfer challenges in evolving AI infrastructure.