Hyphastructure unveils edge cloud for real-time AI at the edge

Hyphastructure has launched what it describes as the world's first distributed edge cloud network designed specifically for physical AI and powered by Intel Gaudi 3 AI accelerators.

The new platform places data centre-grade infrastructure directly where data is generated, at the physical edge, and is engineered to deliver AI inference latency under 10 milliseconds across a network of distributed local nodes.

This capability is intended to enable real-time applications in sectors where fast response times are essential. Among highlighted use cases are autonomous robotics, vehicle-to-vehicle (V2V) collision avoidance, and smart city infrastructure, all of which rely heavily on the minimisation of latency for effective operation.

Architecture and features

The Hyphastructure platform integrates Intel Gaudi 3 AI accelerators, software-defined networking, and bare-metal virtualization into a unified edge solution. This architecture has been designed to allow organisations more flexibility and control in deploying and managing complex AI models directly at the edge.

According to Hyphastructure, the network enables several advantages for customers, including faster AI deployment, improved cost efficiency, and real-time data processing. The company claims enterprises can shift from AI model development to deployment in a matter of hours, in contrast to the weeks that such processes have typically required in the past.

Michael Huerta, Chief Executive Officer of Hyphastructure, said,

"Hyphastructure delivers the infrastructure layer AI-powered industries have been waiting for. By moving inference to the edge, we're removing latency barriers and opening the door to transformative applications from tele-surgery to robotics that were impossible before."

Industry applications

Hyphastructure is aiming the platform at several industry segments. For smart cities, it notes the ability to coordinate infrastructure for traffic, crime, and emergency services directly at the edge. In the retail sector, the platform is positioned to offer real-time shelf monitoring and issuance of personalised offers at the edge, without requiring traditional on-premises resources.

The company also highlights that for autonomous systems and robotics, the network represents the first decentralised platform capable of supporting collision-avoidance inference on V2V networks. This is positioned as essential for the next generation of vehicles, robots, and drones. In the area of gaming and interactive media, Hyphastructure says its sub-10ms edge compute service can support immersive AR and VR experiences beyond the capability of centralised cloud systems due to latency constraints.

Cost and management benefits

The company states that the infrastructure, built with Intel Gaudi 3, provides up to 40% lower total cost of ownership in comparison to traditional GPU-based systems, a figure which it says has been validated by independent benchmarks. The use of software-defined networking and virtualisation aims to ensure seamless orchestration and scalability of AI workloads across a distributed network of nodes, maintaining performance as applications grow or change over time.

By processing data locally, the system is intended to do away with the latency and bandwidth limitations that can be present in centralised cloud arrangements, instead enabling immediate insights and rapid response to data signals as they are received.

Deployment and technology partner

The use of Intel Gaudi 3 AI accelerators is a central part of the hardware layer for the platform. Combined with a software stack designed for dynamic orchestration and bare-metal virtualisation, Hyphastructure targets deployments in industries requiring both high-performance AI inference and reliable low-latency operations at the physical edge.

Hyphastructure's focus is on applications in retail, robotics, healthcare, gaming, and smart urban environments, providing the technological base required for distributed intelligence and rapid-response computing systems.