Hardware / Systems Architect

About this role

SoftQuantus is building the next generation of AI infrastructure. The constraint is no longer models — the constraint is compute efficiency. Most companies are overspending millions on GPU clusters without mastering memory bandwidth optimization, interconnect topology, or performance per watt. This is not a software problem. It is a systems problem. We are looking for a Hardware / Systems Architect to design infrastructure that delivers lower cost per token than hyperscalers.

Focus Areas

• •GPU architecture at low level (CUDA / ROCm)
• •HPC clusters and interconnect topology (NVLink, InfiniBand)
• •Memory systems and bandwidth constraints (HBM)
• •Performance optimization under real production workloads
• •Cost-per-token benchmarking and infrastructure efficiency metrics

Responsibilities

• Design AI compute infrastructure optimized for cost-per-token efficiency
• Architect GPU cluster topologies with optimized interconnect and memory bandwidth
• Define performance baselines and benchmarking methodologies
• Lead hardware selection and vendor evaluation for AI inference deployments
• Collaborate with AI engineering to align infrastructure with model serving requirements
• Build infrastructure that generates revenue, not just research output

Requirements

• Deep expertise in GPU architecture (CUDA / ROCm at low level)
• Strong background in HPC clusters and interconnect systems (NVLink, InfiniBand)
• Solid understanding of memory systems: HBM, bandwidth bottlenecks, cache hierarchy
• Experience with performance profiling and optimization under real workloads
• Systems-level thinking: architecture over abstraction, efficiency over brute force

Nice to Have

• ○Experience with AI inference infrastructure at scale
• ○Familiarity with performance-per-watt optimization for data center environments
• ○Prior work in hyperscaler or cloud provider infrastructure teams
• ○Background in European HPC or sovereign compute initiatives

Skills & Keywords