Network Architect – AI

Dubai / Abu Dhabi / Remote . Full Time | Flexible – 10+ years of high performance networking experience and must have AI data center experience.

Opportunity

DeFiTech is seeking a Network Architect – AI with an excellent, demonstrable background in both AI infrastructure networking and secure enterprise networks. You will design, implement, secure, and optimize the connectivity that supports GPU clusters, distributed AI workloads, enterprise applications, and hybrid cloud environments.

This is a hands-on architecture role for someone who can connect business and workload requirements to practical designs, lead implementation, and resolve difficult production issues. You should be comfortable moving from architecture reviews and capacity models to switch configurations, Linux diagnostics, security policies, and performance testing.

Your work will help provide AI teams with reliable, high-performance infrastructure while protecting enterprise systems, sensitive data, and administrative access. You will work closely with AI and machine learning engineers, cloud and platform teams, cybersecurity specialists, infrastructure engineers, and technology stakeholders.

Strong experience in BOTH AI networks and enterprise secure networks is essential. Candidates must be able to explain their own contribution to real deployments, the design choices they made, the problems they resolved, and the results they measured.

What you will own

You will own network architecture and technical design across AI infrastructure and its enterprise connectivity, from requirements and proof of concept through deployment, acceptance, and operational handover. You will recommend standards, evaluate options, document tradeoffs, and guide implementation in collaboration with the relevant service and security owners.

The scope includes GPU compute fabrics; AI storage, application, and management connectivity; enterprise data center and office networks; secure cloud connectivity; and the network controls that connect these environments safely. The precise platform mix and deployment scale will be discussed during recruitment.

Key Responsibilities

AI network architecture and capacity planning

– Translate training, fine-tuning, inference, data ingestion, and model-serving requirements into network designs with clear capacity, latency, resilience, security, and cost assumptions.

– Design scalable leaf-spine, Clos, or other workload-appropriate topologies. Evaluate oversubscription, bisection bandwidth, port counts, uplink capacity, expansion paths, and failure domains.

– Assess Ethernet with RDMA over Converged Ethernet version 2 (RoCEv2) and InfiniBand options against workload behavior, operational skills, ecosystem compatibility, and total cost of ownership.

– Distinguish GPU scale-up interconnects such as NVLink from scale-out networks between servers. Coordinate GPU, NIC, PCIe, and NUMA topology requirements with compute specialists.

– Plan appropriate separation or controlled convergence of GPU compute, storage, application access, in-band management, and out-of-band management networks.

– Specify switching, network interface cards, optics, cabling, link speeds, and redundancy. Evaluate 100/200/400/800 Gb/s connectivity where justified by the workload and supported platforms.

– Coordinate rack connectivity, fiber and copper choices, transceiver compatibility, cable distances, physical diversity, and network equipment power requirements with data center teams.

– Produce phased growth plans that accommodate additional GPU nodes, storage capacity, tenants, sites, and services without avoidable redesign.

Ownership of:

• AI fabric deployment and performance optimization
• Enterprise network architecture and resilience
• Enterprise security and protection of AI infrastructure
• Hybrid cloud and container networking
• Automation, visibility, and operational readiness
• Architecture governance and technical leadership
  

Expected Experience and Qualifications:

– A minimum of 5 years of relevant professional experience in network engineering, network architecture, data center networking, or closely related infrastructure roles, including substantive design and implementation responsibility.

– Excellent hands-on experience in both AI infrastructure networking and enterprise secure networks, supported by specific deployment examples.

– Practical experience designing, deploying, or materially optimizing a multi-node GPU or AI cluster network. Be prepared to discuss topology, transport, node or GPU scale, link speeds, workload characteristics, security controls, and validation methods.

– Strong practical experience with RoCEv2/Ethernet or InfiniBand fabrics, together with the ability to explain the architectural differences, tradeoffs, and appropriate use cases of both.

– Demonstrated understanding of RDMA, distributed GPU communication, congestion management, host networking, and the relationship between fabric behavior and AI workload performance.

– Strong routing, switching, and troubleshooting capability, including TCP/IP, BGP, OSPF, VLANs, VRFs, redundancy, and modern data center fabric principles.

– Demonstrated experience securing enterprise connectivity through segmentation, firewalls, secure remote and administrative access, authentication integration, and policy validation.

– Practical Linux networking skills and confidence investigating interface state, routing, NIC counters, RDMA devices, firmware and driver compatibility, and host connectivity issues.

– Experience integrating data center or enterprise environments with at least one major public cloud platform, such as AWS, Microsoft Azure, or Google Cloud.

– Working knowledge of container and Kubernetes networking sufficient to design and troubleshoot connectivity with platform engineers and identify policy enforcement boundaries.

– Proven ability to deliver architecture documents, migration plans, acceptance criteria, operational handovers, and clear technical recommendations.

– Strong written and spoken English for design documentation, stakeholder discussions, incident coordination, and technical knowledge transfer.

– Familiarity with NVIDIA networking technologies such as Spectrum-X, Quantum InfiniBand, ConnectX adapters, BlueField DPUs or SuperNICs, and associated management tools.

– Experience with Cisco, Arista, Juniper, or comparable enterprise and data center platforms; open networking and SONiC experience is also valuable.

– Experience with high-speed 400/800 Gb/s AI fabrics, rail-oriented or multiplane designs, and topology choices for larger GPU clusters.

– Exposure to GPU systems and distributed workloads using NCCL, PyTorch, MPI, Slurm, or comparable infrastructure and software components.

– Understanding of storage networking for parallel file systems, object storage, NVMe over Fabrics, or GPUDirect Storage where relevant to the architecture.

– Experience with DPU offload, SR-IOV, RDMA-enabled Kubernetes environments, eBPF-based networking, or advanced workload isolation.

– Experience with accelerated cloud networking, such as AWS Elastic Fabric Adapter, where appropriate to the selected platform.

– Familiarity with network platforms and tools such as NetBox, Terraform, Git-based delivery pipelines, streaming telemetry, Prometheus, Grafana, and vendor fabric management systems.

– Experience designing networks for sensitive data, strong tenant separation, isolated environments, or organizations with demanding security and availability requirements.

– Relevant advanced networking, security, cloud, Kubernetes, or AI infrastructure certifications. Credentials such as CCNP, CCIE, JNCIP, or JNCIE are useful indicators where relevant; certification is not a substitute for demonstrated experience.

The technology names above describe relevant experience. Candidates do not need to have used every listed product. Depth in the essential capabilities, sound engineering judgment, and the ability to work across platforms matter most.

Apply with evidence of what you have built

Send your CV to career@defitech.net and include a brief example of a GPU platform you designed or delivered: the workload, infrastructure scale, your contribution, the hardest constraint, and the result.

Please include two concise, anonymized project examples:

1. An AI network deployment or optimization: your role, cluster or node scale, topology, transport, major performance constraint, actions taken, and measured outcome.

2. A secure enterprise network project: your role, environment, segmentation or access requirements, resilience considerations, implementation approach, and result.

Where possible, distinguish your personal contribution from the wider team’s work. Do not include confidential customer information, credentials, proprietary configurations, or materials you do not have permission to share.