Hardware Systems Architect Jobs (NOW HIRING)

Anthropic trains and serves frontier AI models on some of the largest and most diverse accelerator fleets in the world, and the hardware those models run on is one of the most direct levers we have on capability, cost, and reliability. Our Hardware Systems team owns the system-level architecture of that compute - from the package boundary out through boards, racks, interconnect, power, cooling, and the datacenter interface.

We're hiring senior hardware systems architects and technical leads who can work broadly across the hardware stack and go deep where it's needed, and who have the judgment to make and own directional calls - what to build, what to buy, what to co-design with a partner. You'll set architecture, write the specs that partners and internal teams build against, and own the validation and qualification strategy that proves the result.

The skillsets we need span power, cooling, mechanical, interconnect, signal integrity, data and control plane, system management, reliability and serviceability - applied across networking, compute, storage, and accelerator systems.

You'll work closely with our ML performance, infrastructure software, supply chain, and datacenter teams, and directly with hardware vendors and partners across the stack. And because this is Anthropic, you'll have Claude as a genuine collaborator - we expect this team to push on what AI-assisted hardware design and review can look like.

• Own system path-finding and architecture for your domain across the breadth of compute Anthropic deploys - from concept and requirements through spec, design review, and deployment - and act as a reviewer and thought partner across the others.
• Write and own high-level requirements and specifications - system, board, interface, rack - that partners and internal teams build against.
• Review partner and vendor designs against our requirements, make the build / buy / co-design calls with supply chain and partnership teams, and surface misalignment or schedule risk early enough to act on it.
• Drive the interconnect and networking architecture that ties our systems together - scale-up and scale-out fabrics, NICs and switches, optics, and topology.
• Drive board-, chassis-, and system-level architecture with our partners - major component placement, PCB architecture, connector and cable design, and power and thermal budgeting - at the chassis, rack, and data-hall level.
• Develop performance simulations, hardware models, and "what-if" scenarios to evaluate and choose between candidate hardware architectures.
• Define validation, bring-up, and qualification strategy for new platforms - including the telemetry and reliability targets they need to hit in the fleet - and track the technology roadmaps and vendor landscape that shape what we deploy in the short and long term.
• Partner with ML performance and infrastructure software teams so hardware decisions land well for the workloads that actually run on them.
• Help shape how hardware engineering operates at Anthropic - develop and apply AI-assisted approaches to design, spec review, and validation; establish review forums, spec standards, and partner engagement.

• Deep, hands-on expertise in at least one core hardware-systems domain (e.g., interconnect, power, thermal) for large-scale compute or networking systems.
• Experience owning hardware system architecture at scale - machine, rack, row, and cluster - and authoring specifications and requirements.
• Experience taking hardware from architecture through bring-up and high-volume production deployment.
• Experience working with external hardware vendors and partners - reviewing their designs and holding them to a specification.
• Ability to reason about trade-offs across adjacent hardware domains (e.g., interconnect power thermal mechanical) rather than within a single one.
•  Track record of technical leadership - owning directional decisions and their consequences, and driving alignment across teams and partners.
• Comfort operating with high autonomy and limited process in a fast-moving, ambiguous environment.
• Strong written communication - you'll write specs and reviews that many teams depend on.

• 8+ years in hardware systems architecture or engineering for hyperscale, HPC, AI/ML, or high-end networking platforms.
• Experience with AI accelerator systems (GPU, TPU, or other custom ASIC platforms) and their scale-up / scale-out fabrics.
• Familiarity with the chip-package-system interface and co-design.
• Experience establishing a new function, program, or engineering practice.
• Comfort working with - and curiosity about pushing - AI tools as part of an engineering workflow.
• Degree in EE, CE, ME, CS, or a related field, or equivalent experience.
• Depth in one or more of:
• High-speed SerDes, PCIe/CXL, Ethernet/InfiniBand, and optical interconnect.
• BMC and platform firmware, hardware root of trust, and secure boot.
• 48V/HVDC power delivery and direct liquid cooling.
• Fleet-scale reliability engineering and RAS architecture.
• Signal and power integrity.

• Define the scale-up and scale-out interconnect architecture for a next-generation training platform, from link budget and topology through switch / NIC / optics selection and the bring-up and validation plan.
• Architect the management and control plane for a new hardware system: BMC and host control, power / reset sequencing, secure boot and attestation, and the telemetry surface the fleet needs.
• Own the rack-level power and thermal architecture for a liquid-cooled accelerator rack, and the interface spec between that rack and the datacenter that hosts it.
• Build the reliability model and RAS strategy for a multi-thousand-node fleet - failure budgets, derating policy, telemetry requirements, and serviceability targets - and drive the design changes that hit them.
• Run the technical evaluation of a partner's proposed system design against our requirements, write up where it falls short, and drive the changes end to end.
• Prototype an AI-assisted workflow for partner design review and spec consistency checking, and put it to work on a live platform.