Director of Product, Ecosystem

SambaNova builds the RDU, a dataflow accelerator for AI inference, and delivers it as rack-scale systems for on-premises deployment, as managed capacity in partner data centers (SambaManaged), and as an API (SambaCloud). Our customers are clouds, governments, and enterprises building token factories: data center capacity dedicated to running agentic AI workloads. The RDU's advantage is in decode, and it is increasingly deployed alongside GPUs that handle prefill.

SambaNova earns revenue from that advantage only if three things happen: the right models run on the platform at the right time, the surrounding ecosystem of serving engines, networking, storage, and orchestration treats RDU as a first-class target, and the company concentrates its engineering on the bets with the highest payoff. Core Product is the team responsible for those three things. It decides what SambaNova builds and why, grounds those decisions in benchmarks, competitive intelligence, and unit economics, and partners with engineering to see them through to release.

Responsibilities

SambaNova is hiring a Director of Product, Ecosystem to own ecosystem strategy for our core products across scale-out networking, inference serving, storage and caching, prefill compute, agentic CPU offload, and compiler and RDU research: which integrations we pursue, how we position them, and how partners deploy them. Customers deploy RDU in token factories alongside other vendors' hardware and software. This role decides how RDU integrates at each layer, ships the reference architectures and certifications partners deploy from, and converts that work into demand. It sets the partner strategy that domain product managers and engineering execute within. This individual will be responsible for:

Ecosystem strategy

• Define where RDU integrates into each customer segment's token factory: prefill compute, caching storage, scale-out networking, and agentic CPU offload alongside RDU decode.
• Make explicit build, partner, or ignore decisions for each integration area, write down the reasoning, and revisit it quarterly.
• Anchor positioning on decode economics: tokens per second per user, cost per million tokens, and latency tiers.
• Maintain a versioned ecosystem map with a changelog.

Partner enablement

• Publish joint reference architectures for each integration area, with defined integration points next to RDU decode.
• Own certification of RDU against the common cloud-native stack: Kubernetes, Prometheus, Grafana, and managed-stack vendors.
• Ship operators, metrics exporters, dashboards, and integration documentation as maintained products rather than one-off deliverables.
• Treat enablement as complete only when a cloud provider or partner can deploy from the reference architecture without SambaNova engineering support.

Core software ecosystem

• Secure and deliver co-optimization work with inference-engine builders (vLLM, SGLang, llm-d) and system-software optimizers so their runtimes run efficiently on RDU.
• Represent SambaNova in PyTorch and MLIR community work.
• Connect SambaNova's compiler and RDU research to the external ecosystem so programmability improvements reach the frameworks and toolchains partners build on.
• Support co-engineering with model partners so flagship open models run fastest on RDU.

Demand conversion

• Run joint go-to-market with token-factory partners where the buyer and the data center are shared.
• Own the analyst relationships with Marketing: supply benchmark data to SemiAnalysis and Artificial Analysis and co-develop token-factory reference guidance so buyers' TCO models reflect RDU's decode advantage.
• Feed partner and integration signal into product bets and competitive intelligence, in partnership with the Product Strategy and Competitive Intelligence lead.

Positioning and education

• Own external positioning for the core product surface: how networking, serving, storage, prefill, and offload fit together in a token factory.
• Produce internal educational content on the ecosystem for SambaNova teams.
• Deliver a quarterly ecosystem scorecard to the executive team: partner deployments, reference architectures shipped, benchmark results, and partner-attached pipeline.

How We Measure This Role

We measure this role by partner deployments and partner-attached pipeline. MOUs, press releases, and announced logos do not count until the partner has deployed.

Basic Qualifications

• 10+ years in AI infrastructure, accelerators, or data center systems, including 8+ years in product management with direct ownership of ecosystem, platform, or partner-facing product.
• Shipped partner-facing product artifacts (reference architectures, certification programs, or integration surfaces) that external teams adopted.
• Bachelor's degree in computer science, electrical engineering, or a related technical field, or equivalent experience.

Additional Required Qualifications

• Working fluency across scale-out networking, inference serving, storage and caching, prefill and decode disaggregation, CPU offload, and compiler toolchains, with depth in at least two.
• Track record of converting partnerships into measurable deployments, demand, or revenue.
• Fluency in inference economics: cost per million tokens, tokens per second per user, and latency and throughput tradeoffs by workload.
• Comfortable owning a strategy that other organizations execute. Technical depth in both network fabric topologies and inference serving stacks.
• Concise, quantified written communication for executive audiences.

Preferred Qualifications

• Experience at both a silicon or systems vendor (for example, the CUDA, ROCm, or oneAPI ecosystems, or an inference accelerator startup) and at a model lab or inference provider.
• Has closed and delivered at least one co-optimization engagement with a model provider.
• Contributions to, or close working relationships with, the vLLM, SGLang, llm-d, PyTorch, or MLIR communities.
• Exposure to hardware sales cycles: racks, data center constraints (power, cooling, density), and financing-backed deals.
• Has run joint go-to-market with networking, storage, or compute vendors that share a common buyer.