About the roleYou will design and build the network our inference fleet runs on. Every rack we buy has to be racked, cabled, connected, and turned into capacity the platform can use. The network is what makes that happen. There's no existing network team, no inherited topology, and no reference architecture to copy. You'll build it right the first time, knowing it has to scale site after site.
The problem is also new. The fleet is heterogeneous by design. GPUs handle prefill, and decode runs on accelerators from several ASIC vendors, each with its own interconnect requirements and management tooling. You'll define how those systems connect to each other, to our customers, and to the internet. You'll do it as part of the Data Center Deployment team that stands up every new site.
What you'll do:- Design and own the data center network architecture: the leaf-spine fabric, the high-bandwidth, low-latency paths between GPU prefill and ASIC decode systems, and out-of-band management
- Lead the network build-out for each new site: bill of materials, rack elevations and cabling plans, cross-connects, and bring-up through acceptance testing
- Turn each accelerator vendor's networking requirements into one consistent fleet design
- Set up external connectivity: IP transit, peering, and private connections to customers
- Build tenant isolation and network security in from day one, not bolted on after the first enterprise customer asks
- Automate configuration, provisioning, and validation against a real source of truth, so site five goes up faster than site one
- Build monitoring and telemetry for the fabric, and own network incident response
- Manage network hardware vendors, carriers, colocation providers, and remote-hands teams
- Work at the boundary with the platform team. Own the physical and L2/L3 network, fabric performance, and site connectivity. Hand off to the Platform Engineer for fleet-wide request routing, model placement, and scheduling rather than owning that layer yourself
What we need from you:- Hands-on experience designing and deploying data center networks, not only operating existing ones
- Deep routing and switching knowledge for modern DC fabrics: BGP, EVPN-VXLAN or equivalent, ECMP, leaf-spine design
- Experience building high-performance networks for AI/HPC clusters or other latency-critical environments (RDMA/RoCE, InfiniBand, congestion control, 400G/800G optics)
- Network automation skills (Python, Ansible, or similar) and comfort treating network config as code
- Comfort on the data center floor: working through cabling, optics, and hardware failures yourself or by guiding remote hands
- Comfort with hardware heterogeneity and ambiguity
- Genuine interest in being an early hire at a ~6-7 person company
Nice-to-haves:- Networking experience with non-NVIDIA accelerators (SambaNova, Cerebras, TPU pods, or similar)
- Multi-tenant cloud networking and tenant isolation at scale
- Experience with transit contracts, peering, and internet exchanges
- Network source-of-truth tooling (NetBox, Nautobot) and streaming telemetry (gNMI)
