AI/HPC Consultant - London - £837.26 Per Day Umbrella

We are recruiting for a AI/HPC Consultant for a leading IT service provider based in London.

The AI/HPC Consultant will support the definition, design and validation of advanced AI infrastructure solutions based on NVIDIA reference architectures. The role is heavily focused on high-performance networking and physical connectivity design across GPU-accelerated platforms, with regular involvement in bill-of-materials reviews, design assurance and customer-facing technical guidance.

Typical activities

- Validate NVIDIA reference architecture choices against customer scale, rack layout, cabling distances, data centre constraints and growth expectations.

- Review proposed switch, optic and cable selections to confirm correct speed, form factor, reach, fibre type, connector type and supported use case.

- Compare design options such as InfiniBand versus Spectrum-X Ethernet Back End, MMF versus SMF, DAC versus optical and centralised versus distributed switch placement.

- Produce clear diagrams, port maps, cable matrices, BoM annotations and design assumptions to support sales, delivery and customer review.

- Engage with NVIDIA and OEM partners to validate emerging design patterns, component availability and product-specific constraints.

- Support handover into delivery, including explanation of design decisions, known constraints, dependencies, risks and operational considerations.

- Desirable certifications and background

- NVIDIA certifications or demonstrable equivalent experience in AI infrastructure, networking or data centre technologies.

- Cisco CCNP/CCIE, NVIDIA Networking, HPE, Aruba, Juniper, Arista or other relevant data centre networking certifications.

- Experience in professional services, consultancy, pre-sales architecture or customer-facing technical design roles.

- Experience with high-performance storage platforms used in AI/HPC environments.

- Familiarity with automation, monitoring, observability or infrastructure-as-code tooling used to deploy and operate data centre fabrics.

Essential experience and skills

- Strong networking background with demonstrable experience in data centre, HPC, AI infrastructure or large-scale low-latency fabric design.

- Excellent knowledge of high-performance AI/HPC networking technologies, including InfiniBand, RoCEv2, Spectrum-X Ethernet, congestion control concepts and lossless Ethernet design principles.

- Strong understanding of VXLAN EVPN, BGP, leaf-spine fabric design, routing design, resiliency models and operational troubleshooting in modern data centre networks.

- Good understanding of physical infrastructure design, including switch placement, row-level design, structured cabling, patching strategy, fibre polarity, cable routing and data centre implementation constraints.

- Practical knowledge of optical and copper connectivity choices, including MMF, SMF, DAC, ACC/AEC, breakout cables, MPO/MTP, LC, QSFP, QSFP112, OSFP and QSFP-DD form factors.

- Ability to interpret and validate NVIDIA reference architectures and translate them into customer-specific designs and commercially accurate Bills of Material.

- Strong written communication skills, including the ability to produce clear HLDs, LLDs, design notes, BoM justifications and customer-facing technical explanations.

- Confident stakeholder engagement skills, including working with customers, partners, vendors and internal delivery teams.

- Analytical mindset with the ability to spot design risks, interoperability issues, unnecessary cost, unavailable components or avoidable cabling complexity.

Technical capabilities

- NVIDIA AI/HPC platforms

- GB300, B300, DGX, HGX, NVL72, GPU node connectivity, ConnectX, BlueField/SuperNIC concepts and NVIDIA reference architecture interpretations

- InfiniBand Back End fabrics

- NDR/XDR/HDR concepts, rail-optimised design, spine/leaf sizing, congestion and resiliency considerations, and Quantum-class switching such as QM3400.

- Spectrum-X Ethernet

- Spectrum-X Front End and Back End use cases, RoCEv2, multi-plane designs, Spectrum switching such as SN5610, Ethernet optics and cabling patterns.

- Data centre networking

- Leaf-spine design, VXLAN EVPN, BGP, resilient routing, underlay/overlay separation, management, storage and in-band/out-of-band network segmentation.

- Optics and cabling

- MMF versus SMF, DR4/FR4/LR4, SR4, DAC/ACC/AEC, breakout/splitter assemblies, MPO-12/APC, LC duplex, reach limits, form factors and compatibility risks.

- BoM optimisation

- Ability to identify practical cost, availability and implementation improvements without compromising performance, supportability or reference architecture alignment.

Essential

- Physical infrastructure

- Rack/row placement, switch placement, cable length planning, patching, cooling impact, power density and layout considerations for high-density AI clusters.

Strongly preferred

- Liquid cooling

- Awareness of direct liquid cooling, in-row versus in-rack CDU options, operational constraints and how cooling decisions affect rack layout and networking design.

Desirable

- Storage and management integration

- Understanding of high-speed storage connectivity, management fabrics, OOB networks and how they integrate with AI infrastructure designs.

This is an umbrella contract, the role is Inside IR35