NexGen Cloud - Technical Architect

• Rather than a checklist, here’s what success in this role actually looks like: • Take each cluster from CTO handover to customer acceptance as its single design authority, so nothing is ordered or installed against a design you haven’t validated • Turn high-level designs and initial BOMs into build-ready low-level designs, hall layouts, rack elevations, P2P maps and cable schedules that site teams can build from without having to call you • Design the InfiniBand and Ethernet fabric, power, direct liquid cooling and containment to fit the real constraints of each data centre • Validate every BOM before it goes to procurement, and make part substitution calls knowing exactly what breaks when a line changes • Hold OEMs and integrators (NVIDIA, Dell, Lenovo, Supermicro and others) to their technical commitments, from FAT witness through to on-site issue resolution • Build the reference designs and building blocks that mean each new deployment starts from a known-good baseline rather than a blank page • We care more about how you think and work than a perfect CV. You’ll likely bring a combination of the following: • You’ve designed and delivered large-scale GPU or HPC clusters into live data centres, and you were on site for the build • Deep, practical knowledge of NVIDIA platforms (GB200/GB300 rack-scale and HGX H200/B200) and what each one demands of a facility • Cluster-scale InfiniBand and Ethernet fabric design: topology, rail optimisation, oversubscription, and cable and optic selection • Physical infrastructure design across power and redundancy, direct liquid cooling and CDUs, containment and achievable cable lengths • You know part numbers, what’s optional and what isn’t, and you can hold an OEM to account when their commitments slip • Factory integration and FAT experience, including holding a dispatch gate • Storage design for AI workloads (WEKA, VAST, DDN, Ceph or similar) • Deployments across multiple geographies and data centre operators, including sites requiring fit-out • Neocloud or hyperscale experience on compressed schedules