Hedgehog Opens AI Networks with OCP, Challenging Vendor Lock-In

BARCELONA, SPAIN – April 30, 2026 – In a move poised to accelerate the shift toward open standards in artificial intelligence, AI network company Hedgehog today announced it has contributed its AI training and inference fabric designs to the Open Compute Project (OCP). The designs, which have achieved OCP Accepted™ status, provide production-ready blueprints for building high-performance, Ethernet-based AI infrastructure and are now available on the OCP Marketplace.

The announcement, made at the 2026 OCP EMEA Summit, signals a significant push against the proprietary, closed ecosystems that have long dominated high-performance computing. By open-sourcing validated network architectures, Hedgehog aims to lower the barrier to entry for companies building powerful AI clusters, allowing them to avoid vendor lock-in and reduce complexity.

"Our goal has always been to make AI networks easier to deploy and operate in the real world," said Marc Austin, CEO of Hedgehog. "By contributing these AI training and inference fabrics to OCP, we're sharing already proven designs that help the community move faster while preserving choice across hardware and silicon."

Democratizing AI with Open, Cost-Effective Blueprints

The core challenge for many organizations racing to deploy AI is the prohibitive cost and complexity of the underlying infrastructure. Specialized networking, often tied to a single vendor's hardware and software stack, can consume a substantial portion of an AI cluster's budget. Hedgehog's contribution directly targets this issue by championing an open, disaggregated model using standards-based Ethernet.

This approach has already demonstrated significant real-world benefits. FarmGPU, a GPU cloud provider, reported that by using an OCP-based solution with Celestica hardware and Hedgehog's open networking software, they built a backend fabric for their GPU clusters that was approximately 50% cheaper than a comparable proprietary InfiniBand architecture. This cost saving allows companies to redirect capital toward what matters most: computational power in the form of more GPUs.

"Our long-term GPU customers need a network that keeps their clusters fed — no bottlenecks, no surprises," said Jonmichael Hands, CEO of FarmGPU. "Hedgehog made our backend fabric setup straightforward and their support team has been rock solid. Getting these designs into OCP means more operators can run real AI infrastructure without reinventing the wheel."

The reference architectures are split into two distinct designs tailored for specific AI workloads:

• AI Training Fabrics: Engineered for massive, large-scale GPU clusters, these designs prioritize predictable performance. They incorporate features like congestion-aware routing and lossless Ethernet to ensure the high-bandwidth, uninterrupted data flow essential for training large models efficiently.
• AI Inference Fabrics: Optimized for deploying trained models, these architectures focus on efficiency, low latency, and multi-tenant security. They are built for consistent performance at scale, enabling real-time AI applications.

The Technical Shift to Ethernet for AI Dominance

For years, InfiniBand was the de facto standard for high-performance computing and AI networking due to its inherently lossless nature. However, the industry is now undergoing a seismic shift toward Ethernet. Driven by hyperscalers and a vast, multi-vendor ecosystem, modern Ethernet can now deliver the lossless, low-latency performance required by demanding AI workloads, but with greater flexibility and cost-effectiveness.

Achieving this performance relies on advanced standards like RDMA over Converged Ethernet (RoCEv2), Explicit Congestion Notification (ECN), and Priority Flow Control (PFC). Hedgehog's reference architectures leverage these technologies on open hardware, such as Celestica's DS5000 switches powered by Broadcom's Tomahawk 5 silicon, to create a robust network fabric. This approach effectively mitigates network congestion—a critical problem in AI training where massive "elephant flows" of data can choke a network and leave expensive GPUs idle.

This ecosystem-driven collaboration is key to validating the designs. As a Platinum OCP Member, hardware partner Celestica worked closely with Hedgehog to ensure the architectures were proven across open systems.

"These reference architectures show how Celestica's OCP-inspired switches and open source networking software can be combined to support modern AI workloads," said Olivier Suinat, Chief Revenue Officer, Enterprise AI Platforms at Celestica. "Celestica is proud to support contributions like these that give customers a clear, deployable path to building open and scalable AI infrastructure aligned with OCP standards."

Solidifying OCP's Role in the Future of AI

Hedgehog's contribution is a key part of a larger trend: the Open Compute Project's expanding influence in shaping the future of AI infrastructure. Originally focused on standardizing traditional data center hardware like racks and servers, OCP has strategically pivoted to address the monumental demands of AI. Its "Open Systems for AI" initiative is now a focal point for industry collaboration on everything from advanced liquid cooling to high-performance networking.

The market is responding with force. According to a recent IDC forecast, total spending on OCP-recognized IT infrastructure is projected to reach $295 billion by 2029, with AI infrastructure deployment being a primary driver. This explosive growth underscores the industry's appetite for the efficiency, scalability, and supply chain diversity that open standards provide.

By publishing validated reference architectures, OCP provides a trusted foundation for operators and system builders. This reduces the integration risk and debugging time that can sometimes accompany open, disaggregated systems, giving organizations the confidence to adopt them.

"As AI workloads drive new demands on data center networks, the OCP Community is focused on enabling open, Ethernet-based solutions that can scale efficiently and operate reliably in production," commented James Kelly, VP of Market Intelligence and Innovation for the Open Compute Project Foundation. "By publishing these validated AI fabric reference architectures... we are giving operators and system builders direct access to designs that make it easier to adopt open, disaggregated AI networking with confidence and clarity."

This convergence of open software from companies like Hedgehog, flexible hardware from partners like Celestica, and the standard-setting influence of OCP is creating a powerful alternative to closed ecosystems. As AI continues its exponential growth, these open, collaborative efforts are becoming essential for building the next generation of data centers in a sustainable and scalable way.