Hedgehog Contributes Open Compute Venture Reference Architectures for AI Coaching and Inference

New OCP Accepted™ designs speed up deployment of open, scalable AI infrastructure

Hedgehog, the AI community firm simplifying how AI infrastructure is constructed and operated, introduced it has contributed its AI coaching material and AI inference material designs to the Open Compute Venture (OCP) as reference architectures, and are instantly out there by way of the OCP Market. The OCP Accepted™ recognition applies to merchandise that align/adjust to accepted OCP specs

Introduced along side the 2026 OCP EMEA Summit, going down in Barcelona, Spain April 29 – 30, the contributions present operators, system builders, and integrators with validated, production-ready blueprints for deploying open, Ethernet-based AI networks utilizing disaggregated {hardware} and Hedgehog AI community software program.

Our objective has all the time been to make AI networks simpler to deploy and function in the true world.

Open AI Materials Constructed for Actual Deployments

Primarily based on real-world manufacturing deployments supporting as we speak’s most demanding workloads, The reference architectures emphasize interoperability throughout silicon distributors to forestall {hardware} lock-in. The designs tackle distinct atmosphere necessities:

• AI Coaching Materials: Designed for large-scale GPU clusters, delivering predictable efficiency by way of congestion-aware routing, lossless Ethernet, and automatic community lifecycle administration.
• AI Inference Materials: Optimized for effectivity and low latency, that includes multi-tenant safety, hybrid multi-cloud routing, simplified operations, and constant efficiency at scale.

“Our objective has all the time been to make AI networks simpler to deploy and function in the true world,” mentioned Marc Austin, CEO, Hedgehog. “By contributing these AI coaching and inference materials to OCP, we’re sharing already confirmed designs that assist the group transfer sooner whereas preserving selection throughout {hardware} and silicon.”

Additionally Learn: AiThority Interview with Glenn Jocher, Founder & CEO, Ultralytics

Ecosystem Collaboration from Design to Manufacturing

The reference architectures mirror shut collaboration throughout the OCP ecosystem, previewed initially throughout a joint panel on the OCP International Summit 2025, the place a number of firms shared classes realized from deploying Ethernet-based AI materials at scale.

“Our long-term GPU prospects want a community that retains their clusters fed — no bottlenecks, no surprises,” mentioned Jonmichael Arms, CEO, FarmGPU. “Hedgehog made our backend material setup simple and their assist workforce has been rock stable. Getting these designs into OCP means extra operators can run actual AI infrastructure with out reinventing the wheel.”

As an OCP {hardware} associate, Celestica labored intently with Hedgehog to validate the architectures throughout open techniques.

“These reference architectures present how Celestica’s OCP-inspired switches and open supply networking software program may be mixed to assist fashionable AI workloads,” mentioned Olivier Suinat,Chief Income Officer, Enterprise AI Platforms, Celestica. “Celestica is a Platinum OCP Member and proud to assist contributions like these that give prospects a clear, deployable path to constructing open and scalable AI infrastructure aligned with OCP requirements.”

Advancing Open AI Infrastructure with OCP

The contribution aligns immediately with OCP’s Networking Venture to advance scalable, high-performance infrastructure.

“As AI workloads drive new calls for on information middle networks, the OCP Neighborhood is concentrated on enabling open, Ethernet-based options that may scale effectively and function reliably in manufacturing,” mentioned James Kelly, VP Market Intelligence and innovation, Open Compute Venture Basis. “By publishing these validated AI material reference architectures by way of the OCP Contributions Database and related merchandise within the OCP Market, we’re giving operators and system builders direct entry to designs that make it simpler to undertake open, disaggregated AI networking with confidence and readability.”