HPC Network Architect for AI Clusters

CommonAI Holdings Ltd; Cambridge; Hybrid; GBP 70,000 - 110,000; Full time

Benefits offered by this job

Competitive salary package with a UK-p

Hybrid/remote working option

Cambridge office near train station

Job summary

CommonAI CIC in Cambridge seeks a senior network architect to design and deliver the physical networks for our Scaling Inference Lab. You will own cabling, topology, switch layout and BOMs across out-of-band, frontend/storage and accelerator fabrics, coordinating with vendors and internal teams.

Experience with InfiniBand, RoCE, OSFP/QSFP-DD optics and dense cabling is essential. You will communicate designs to installers and project managers, support bring‑up and fault‑finding, and travel

Qualifications

• Experience designing networks for HPC or AI cluster deployments, with examples of designs you have owned from specification to installation.
• A thorough understanding of the differences between out-of-band, frontend and accelerator (GPU) fabric networking.
• Practical knowledge of InfiniBand, RoCE or similar lossless fabrics, including fat-tree and rail-optimised topologies.
• A very clear understanding of data centre fibre optics: OS2 versus OM4/OM5, MTP/MPO versus LC, APC versus UPC, Type A/B/C polarity, pinned and unpinned connectors, and the losses introduced by patch panels, cassettes and connectors.
• The ability to communicate a design clearly to installers, suppliers, project managers and engineers.
• Experience with 400G/800G Ethernet and OSFP/QSFP-DD optics, optical circuit switching or other novel interconnects, high-density liquid- or air-cooled racks, documentation tools such as NetBox, and colocation facilities would be useful. Depth in a subset is sufficient; the lab will keep presenting new requirements and we expect you to learn alongside us.
• A collaborative and supportive work environment

Responsibilities

• Design the physical network for each new cluster across out-of-band, frontend/storage and accelerator fabrics: topology, switch placement, port maps, cable schedules and bills of materials.
• Specify fibre and copper infrastructure in detail, including fibre grade, connectors, polarity, pinning, trunk and cassette layouts, patch-panel arrangements and end-to-end loss budgets, and select matching transceivers, DACs and AOCs.
• Plan changes to existing clusters so equipment can be re-cabled and reused with minimal disruption to running work.
• Act as the technical authority for suppliers and installers: produce drawings and specifications, review quotations, and inspect and test completed work.
• Maintain accurate, well-labelled as-built documentation of every rack, cable, port and link.
• Work with project managers on sequencing and lead times, and with operators and software colleagues on bring-up and fault-finding.
• An example early project would be to produce the complete network design for an upcoming cluster, oversee its installation, test the links and hand over as-built documentation to the operations team.
• Experience with 400G/800G Ethernet and OSFP/QSFP-DD optics, optical circuit switching or other novel interconnects, high-density liquid- or air-cooled racks, documentation tools such as NetBox, and colocation facilities would be useful. Depth in a subset is sufficient; the lab will keep presenting new requirements and we expect you to learn alongside us.
• A collaborative and supportive work environment

Skills

HPC network design

AI cluster networks

InfiniBand

RoCE

Data centre fibre optics

Clear design communication

Tools

NetBox

OSFP/QSFP-DD optics

Job description

CommonAI CIC in Cambridge seeks a senior network architect to design and deliver the physical networks for our Scaling Inference Lab. You will own cabling, topology, switch layout and BOMs across out-of-band, frontend/storage and accelerator fabrics, coordinating with vendors and internal teams.

Experience with InfiniBand, RoCE, OSFP/QSFP-DD optics and dense cabling is essential. You will communicate designs to installers and project managers, support bring‑up and fault‑finding, and travel