Head of Compute and Capacity - Lithos Ai

Unusual Ventures • San Francisco, CA

7d

About The Position

Lithos builds fast agentic inference using frontier open models at full precision on standard GPUs. The company serves startups and Fortune 500 companies through a public API and on customers’ own hardware. Founded by Carnegie Mellon computer science professors, Lithos is a small team, and this role is one of its first ten hires. The Head of Compute and Capacity owns the company’s compute capacity strategy, which is central to delivering speed, price, and uptime commitments. This is described as Lithos’s most important commercial role after the founders, reporting directly to them.

The role is responsible for translating demand forecasts and the model roadmap into multi-year GPU, network, and data-center requirements; procuring capacity across vendors; negotiating commercial terms; tracking vendor performance and market pricing; modeling total cost of ownership; and partnering with engineering to ensure purchased hardware performs well with Lithos’s software. The person will take over live, in-flight deals on day one and initially operate as a team of one, building the necessary processes, spreadsheets, and vendor relationships.

Requirements

• At least 7 years of experience in compute capacity, infrastructure procurement, or data-center commercial roles at a neocloud, colocation provider, OEM or channel partner, or hyperscaler capacity team.
• Personally negotiated and closed infrastructure contracts worth seven figures or more, including commit structures, drawdowns, and SLA remedies, and be able to explain one in detail.
• Ability to reason and communicate in quantitative terms, including cost per GPU-hour, utilization, power density, delivery lead times, and effective cost per unit of performance.
• Working knowledge of current GPU generations and high-speed networking technologies such as InfiniBand and RoCE, including the factors that distinguish a good cluster from an expensive one.
• Comfort operating initially as a team of one and building spreadsheets, processes, and vendor relationships independently.

Nice To Haves

• Experience managing capacity for AI inference or training workloads.
• Experience on the vendor side selling GPU capacity to AI labs.

Responsibilities

• Own the multi-year capacity plan, translating demand forecasts and the model roadmap into GPU, networking, and data-center requirements while anticipating demand.
• Source and negotiate GPU capacity, networking, and colocation across neoclouds, colocation providers, OEMs, and hyperscalers; build a diversified vendor bench.
• Lead infrastructure contracts end to end, including commit profiles, drawdown schedules, pricing, acceptance testing, uptime definitions, service credits, network fabric performance guarantees, and SLA remedies.
• Take over live, in-flight deals on the first day.
• Build and maintain a rolling price book and vendor scorecard covering market pricing, delivery reliability, and real-world performance by vendor, region, and GPU generation.
• Model total cost of ownership per unit of performance, including cost per token and cost per GPU-hour based on actual hardware performance, and use this analysis for capacity decisions.
• Partner with systems and inference engineering teams to ensure purchased hardware suits the software and that acceptance tests verify its performance.
• Initially build the spreadsheets, processes, and vendor relationships as a team of one.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: No Education Listed