Hardware System Architect Lead, GPU Platform

About Think AI

The Role

Responsibilities

• Own system architecture, PCIe topology, switch selection and partitioning, lane budgets, and GPU-to-NIC affinity.
• Design and validate GPUDirect RDMA and peer-to-peer traffic and vendor-specific ACS, ATS, IOMMU, and P2P behavior.
• Own signal-integrity strategy and lead power and thermal architecture for dense accelerator systems.
• Define BIOS/UEFI and BMC requirements and coordinate hardware integration with the ILM software team.
• Write platform specifications and reference designs; review schematics, layouts, and BOMs.
• Manage ODM/JDM partners and drive platforms through EVT, DVT, and PVT into Riyadh production.
• Transition board design in-house and guide future generations, including Gen6 switching, CXL, and rack-scale integration.
• Own validation, qualification, design-for-test, factory test coverage, burn-in, and yield targets.
• Lead failure analysis and corrective actions; build Python and Bash test automation and grow the hardware platform team.

What You Bring

• 12+ years in hardware or system architecture focused on GPU-accelerated or accelerator-dense servers.
• Have shipped at least one dense multi-GPU server platform from architecture through PVT and production.
• Deep PCIe expertise, including switch topologies, Gen5/Gen6 signal integrity, retimers, enumeration, and BAR/resource allocation.
• Strong working knowledge of RDMA NICs, GPUDirect, NUMA, IOMMU, GPU-to-NIC communication, and power and thermal constraints.
• Able to read schematics, lead layout reviews, and debug with a scope and PCIe protocol analyzer.
• Hands-on platform bring-up, validation, and failure analysis through production ramp, including factory debug.
• Experience with validation plans, qualification matrices, root cause and corrective action, and Python and Bash test automation.
• Fluent English.

Nice to Have

• Hands-on schematic design for server motherboards, risers, or PCIe switch and retimer boards.
• Experience with multiple accelerator vendors on one platform; UALink, Ultra Ethernet, NVLink-class interconnects, or CXL.
• Experience with HGX, OAM, UBB, OCP/ORv3, liquid-cooled servers, DDR5 qualification, rack-level integration, HALT, or HASS.
• Startup experience scaling from early to growth stage.

What Think Offers

• Early-stage compensation and meaningful ownership and impact at an early-stage deep-tech company.
• Close collaboration with a small, senior team on hardware, systems, and AI problems at scale.