About the company
Our client is a venture-backed energy technology company building a new kind of AI compute network. Instead of one large data center, they deploy liquid-cooled NVIDIA GPU inference pods across an established network of more than a thousand properties in well over 100 US metros, running on permitted, grid-connected power that is already in place.
The pods are designed to be modular and quick to deploy. They avoid new utility permits, water hookups, backup generators and major construction. Each pod shares its site's electrical service with other energy infrastructure, so the platform balances power intelligently and fails over gracefully between sites.
The company is now scaling from its first pod to metro-wide fleets and, over time, thousands of distributed inference sites.
The role
Now seeking a talented Head of AI Infrastructure. This is the company's first dedicated infrastructure hire, reporting directly to the CTO.
You'll start as a senior individual contributor, hands-on and owning the AI platform end to end: compute, memory, network, storage, orchestration and field operations. You'll also be the senior technical voice for customers and set the 12 to 18 month infrastructure roadmap.
The team
Over the first 6 to 12 months, as pods deploy across metros, you'll hire engineers across platform/SRE, networking, storage and field operations, then add managers as the fleet grows.
Cloud infrastructure, IT and security sit with the VP of IT & Security. Properties and installation, and vendor management, have their own teams. You'll partner closely with all three.
What you'll do
- Bring up, burn in and run GPU pods in production, and own the acceptance benchmarks every new pod and hardware generation must pass
- Measure and improve inference performance across compute, memory, network and storage: tokens per second per kW, KV-cache offload, RoCEv2 and NCCL fabric performance, model cold-start
- Test and roll out frequent driver, vLLM and model updates safely, with clear benchmarks and rollback plans
- Run power-aware operations: GPU power caps, curtailment and workload drain coordinated with other on-site energy loads
- Design for graceful failure, so work hands off cleanly between pods and sites
- Operate a multi-tenant, bare-metal Kubernetes GPU platform against service level objectives, with 24/7 incident response
- Write the runbooks field technicians follow at unmanned sites
- Act as technical lead for customers, and set the 12 to 18 month infrastructure roadmap
- Hire and lead a team across platform/SRE, networking, storage and field operations as the fleet scales
What you'll bring
- 12+ years in infrastructure, with hands-on ownership of physical production compute at scale (not only consuming public cloud services)
- Experience at a hyperscaler, GPU cloud or neocloud: you know how large organizations keep large systems running
- Depth in at least two of: inference serving (such as vLLM), InfiniBand or RoCE networking, distributed storage, bare-metal Kubernetes
- Strong Linux skills, plus Python or Go: you read the code and write the fix
- Happy to start as a hands-on individual contributor, with the ambition to build and lead a team
Nice to have
- Large-scale GPU cluster management (1,000+ GPUs)
- Power-aware scheduling, GPU power capping or demand response
- Edge infrastructure, such as CDN points of presence, cloud local zones or telecom edge
- Building AI compute infrastructure from the ground up
- Background at a GPU, chip or infrastructure vendor, such as NVIDIA, AMD, Intel or Oracle
You don't need to check every box to be considered.
Why join
- Permitted, grid-connected power is already in place across a nationwide property network, so pods deploy in weeks rather than years
- The first infrastructure hire, reporting to the CTO, shaping the platform and the team from day one
- Travel: occasional travel to pod sites and customers
- Work authorization: must already be authorized to work in the US or Canada; visa sponsorship is not available
- Compensation: competitive base salary, bonus and equity included