This deployment signals a growing trend of infrastructure providers diversifying away from single-vendor GPU strategies, with ROCm enabling enterprise migrations previously blocked by framework compatibility concerns. The 110MW capacity commitment from Bitzero provides revenue visibility that greenfield AI colocation operators lack.
- AMD Instinct MI355X (CDNA 3) providing 1.7 TB/s memory bandwidth per node in initial rollout
- ROCm 6.x stack validated for PyTorch 2.
Partnership pairs MI355X GPUs and ROCm stack with OneQode's existing telco footprint
OneQode has announced a strategic infrastructure deployment powered by AMD Instinct GPUs, targeting AI workloads across Europe and Asia-Pacific markets. The initial phase centers on AMD Instinct MI355X GPUs running the open ROCm software stack, with AMD Helios rack-scale solutions integrated into future expansion phases.
The announcement lands amid surging demand for sovereign AI infrastructure. OneQode's recent 110MW capacity agreement with Norwegian operator Bitzero provides the foundational load for this deployment. The company brings seven years of cloud and telecommunications presence across five continents, enabling a shorter ramp to revenue compared to greenfield infrastructure plays.
"We needed a vendor-neutral foundation that wouldn't lock our enterprise and government customers into proprietary training ecosystems," a OneQode technical lead stated. "ROCm's support matrix for major frameworks gave us confidence the migration path stays open."
Technical Architecture
The MI355X deployment leverages AMD's CDNA 3 architecture, delivering 1.7 TB/s memory bandwidth per node. For large-scale training runs, the Helios platform scales to multi-rack configurations with AMD Infinity Fabric interconnects maintaining coherent memory access across nodes. OneQode engineers confirmed the ROCm 6.x stack supports PyTorch 2.2 and JAX natively, reducing the custom kernel work that typically plagues GPU migrations.
For customers requiring data residency, OneQode's sovereign AI tier offers dedicated bare-metal clusters with geographic pinning. The company's telco partnerships provide low-latency backbone access, targeting inference workloads under 20ms round-trip in major European metro areas.
Market Positioning
The infrastructure targets three distinct segments: frontier model training requiring multi-GPU clusters, enterprise AI deployments migrating from NVIDIA-dominated environments, and government research workloads with compliance constraints. OneQode expects Helios integration to accelerate availability of larger training runs, with rack-scale deployment at Norwegian facilities planned for Q3.
Competitors in the European AI infrastructure space include Crusoe Energy, which recently expanded Icelandic facilities, and Scale Wave's German operations. OneQode's telco heritage provides differentiation for latency-sensitive inference applications, though the ROCm adoption signals a broader shift toward vendor-neutral infrastructure strategies across the hyperscaler tier.
The deployment enters initial testing phases next quarter, with commercial availability tied to ROCm 6.2 validation cycles.
---
M4S TAKE
My take: partnerships only work when both sides bring something the other cannot build quickly. The test is whether the combined offering solves a problem neither could address alone. If it does, this is worth watching.
Simon McLoughlin
Is this your company?
This article features your business. Claim it to add your logo, contact details, and a link to your website — or upgrade to reach more buyers.
Did you know 80% of Press Releases trigger AI content warnings? Reach out and the M4S team can assist.