Network Architect

Together AI•San Francisco, CA

31d•$250,000 - $280,000

About The Position

Together AI is building the next-generation AI compute platform, and networking is at the center of that mission. As a Network Architect, you will define and evolve the global network architecture that powers our AI training, inference, and research platforms. This is a deeply technical and strategic role: you will own the end-to-end routing, topology, traffic engineering, and control-plane strategy for a global network spanning self-built data centers, partner colo, cloud environments, and high-capacity backbone fabrics. You will collaborate closely with infrastructure engineering, compute systems, hardware, and operations teams to design architectures that deliver massive east-west bandwidth, low latency, high resiliency, and predictable performance at multi-terabit scale. Your work directly influences how we build, scale, and operate the physical and logical networks that underpin cutting-edge AI workloads. This is a role for architects who are hands-on enough to validate designs in production, experienced enough to reason about systems at huge scale, and creative enough to develop architectures that don't exist yet.

Requirements

Have deep experience designing and operating large-scale GPU clusters or HPC-style compute fabrics, and understand the unique demands these workloads place on network design (east-west dominance, congestion behavior, fan-in/fan-out patterns, loss sensitivity).
Are fluent in building high-throughput data center fabrics (leaf-spine/superspine/core) that support tens of thousands of GPUs, multi-terabit east-west traffic, and strict performance SLAs.
Have architected or operated RoCEv2 or lossless Ethernet environments at scale-including PFC/ECN tuning, congestion control, and end-to-end stability considerations.
Are experienced designing backbone and DCI architectures that support GPU training clusters across multiple regions, interconnect exotic fabrics, and handle high-volume synchronization traffic.
Have led architecture for networks spanning multiple clouds, private backbones, and diverse PoPs, and understand how AI workloads behave across these domains.
Design with operational realities in mind: observability, capacity modeling, automation, telemetry, and failure-mode analysis for GPU-heavy environments.
Are comfortable setting architectural direction in fast-moving environments where compute, storage, and network evolution are tightly coupled.

Responsibilities

Define and evolve Together AI's global routing and backbone architecture, spanning self-built data centers, partner colocation sites, PoPs, cloud regions, and interconnect fabrics.
Establish the end-to-end topology strategy for high-bandwidth AI workloads: east-west fabrics, spine/superspine/core, DCI, and cross-region interconnect.
Design traffic engineering, load balancing, and capacity planning models to ensure low latency, deterministic performance, and fault tolerance at scale.
Develop the multicloud interconnect and peering strategy, including BGP policy frameworks, route leak mitigation, and security posture across heterogeneous networks.
Architect the control-plane stack for programmability, stability, and automation-including routing design, provisioning, configuration management, and state consistency.
Establish foundational observability primitives for a global backbone (telemetry, flow sampling, path validation, synthetic testing, health models).
Work closely with compute, storage, hardware, and data platform teams to ensure network design meets the performance demands of distributed AI training workloads.
Collaborate with operations and NOC teams to ensure designs are supportable, debuggable, and resilient under real-world failure conditions.
Provide architectural direction and mentorship to engineers across the org, influencing long-term strategy for both physical and virtual network domains.
Model evolving topologies for next-generation workloads (multi-Tbps east-west, high fan-in/fan-out distributed systems, GPU cluster fabrics).
Evaluate and guide the adoption of emerging technologies: advanced optical transport, RoCEv2, high-speed Ethernet fabrics, Infiniband overlays, EVPN/VXLAN, SR-MPLS/SRv6, programmable data planes, and hardware offload.