Principal AI Infrastructure Architect

About Nava 

Nava is a neocloud company purpose-built for the AI era. We design, deploy, and operate large-scale GPU infrastructure and deliver inference-as-a-service to teams building the next generation of AI products. Our platform runs on NVIDIA GPU systems, high-performance RDMA fabrics, and a fully automated, software-defined operations model. Every engineer at Nava works close to the metal, on infrastructure built to keep GPUs saturated and models serving.

About the Role 

This is a principal individual-contributor role on the Architecture function. You architect coherent, deployable GPU cluster designs across compute, network, storage, management and physical infrastructure constraints. 

Architect end-to-end AI infrastructure integrating GPU compute, high-performance networking, distributed storage, and management systems. Define reference architectures, capacity models, performance standards, resiliency requirements, and scalable cluster deployment designs. 

What You’ll Do 

• Create reference architectures and low-level designs spanning NVIDIA GPU topologies, high-speed fabrics, storage and management planes. 

• Translate AI workload and scale targets into network bandwidth, storage IOPS/throughput, rack count, power and cooling envelopes. 

• Define topology, failure domains, resilience patterns, lifecycle constraints and expansion strategy. 

• Evaluate trade-offs among NVLink/NVSwitch, InfiniBand/RoCE, parallel storage and orchestration options. 

• Define engineering standards, bill-of-material inputs, performance acceptance tests and design review gates. 

• Validate designs with deployment and SRE teams; drive improvements from field results and benchmark evidence. 

What You’ll Need 

• 7–12 years designing production cloud, HPC or high-performance distributed infrastructure. 

• Deep technical strength in at least two of GPU systems, data center networking, distributed storage or Linux infrastructure. 

• Working grasp of NVIDIA architectures, GPU communication, network fabrics and data center constraints. 

• Ability to produce implementable low-level designs and capacity/performance models. 

• Evidence of architecture decisions deployed and operated in production. 

• Within that, hands-on GPU / AI-datacenter exposure is what we weigh most — we value demonstrated depth over years, as the field is still emerging.