Brian
Yoo

When FriendliAI announced Brian Yoo as its new Chief Business Officer in April 2026, the company's CEO didn't describe a salesman or a networker. Byung-Gon Chun described an engineer of organizations - someone who built "the operational engine behind an AI-driven startup and scaled it into a multi-billion dollar global powerhouse." That company was Moloco. That tenure lasted nearly a decade.

Yoo arrived at Moloco in August 2016 when the company had ten employees and an AI-driven advertising technology that needed a business around it. He left in April 2026 with 600+ people across multiple continents, $250M+ in annual revenue, $180M+ in capital raised, and a valuation hovering around $4 billion. The 500x revenue figure gets cited in press releases, but the more interesting detail is what he actually built to get there: finance, marketing, HR, BizOps, legal, IT, and workplace operations - the entire unglamorous infrastructure of a company, constructed from zero while the product team built the glamorous parts.

That kind of operator doesn't pick a next act casually. Yoo picked FriendliAI, an AI inference platform founded by Seoul National University professor Byung-Gon Chun in 2021, and the choice is pointed. FriendliAI's core proposition is that most enterprises are dramatically overpaying for GPU compute when running large language models in production - and that fixing the inference layer is worth more to AI-driven businesses than almost any product feature they could build.

What FriendliAI Actually Does

The promise sounds like marketing - 2x faster LLM inference, up to 90% GPU cost reduction - but the mechanism is specific. FriendliAI's platform uses custom GPU kernels, continuous batching, speculative decoding, smart caching (the proprietary tcache system), and parallel inference to extract far more throughput from the same hardware. The company can also deploy any of 552,876+ models from Hugging Face Hub in a single click, and it supports multi-LoRA adapters, native quantization, and structured outputs out of the box.

For Yoo, the business logic is clean: inference costs are the largest and least optimized line item in most enterprise AI budgets. Every percentage point of GPU efficiency is real margin recaptured. Every millisecond of latency removed is user experience delivered. The role of CBO is to help enterprises understand that equation and act on it.

The Long Road Through Operations

Yoo's route to this role is not a straight line. Cornell gave him two degrees in Operations Research and Industrial Engineering - a discipline that treats complex systems as optimization problems. He started at Capital One as a business analyst in quantitative risk modeling, then moved to Google's capital markets team doing quantitative analytics. Neither role is on the product-and-fundraising circuit that most startup executives trace. Both are about building rigorous quantitative models for decisions that matter.

Kabam came next, as a Senior Product Manager in mobile gaming from 2013 to 2015. Then a short stint as Chief Strategy Officer at ROOY Inc. before Moloco pulled him in. The arc is someone who spent years learning how businesses actually work before building one - financial modeling, risk analysis, product management, strategy - all disciplines that appear when you have to build a finance team, a legal function, and a global HR operation from scratch without templates.

The Moloco decade also gave Yoo something rarer: credibility inside a technical founder-led AI company. He knows what it takes to build the organizational machinery around a highly technical product. He knows what breaks at 50 people that was fine at 10, and what breaks at 300 that was fine at 100. FriendliAI, with 50 employees and a stated goal of 10x revenue growth in 2026 followed by another 10x in 2027, is about to stress-test those lessons again.

The Inference Bet

The $20M seed extension FriendliAI closed in August 2025 - led by Capstone Partners and joined by Sierra Ventures, Alumni Ventures, KDB, and KB Securities - was meant to fund North American and Asian go-to-market expansion. Yoo is the most visible hire of that expansion. In May 2026, the company opened a 7,000 square foot office in San Francisco's historic Crown Point Press building, signaling that the hiring phase has begun in earnest.

The inference infrastructure market is crowded with technical credibility. The differentiated bet FriendliAI is making - and that Yoo is now the face of commercially - is that efficiency at the serving layer is the AI cost problem most enterprises haven't solved, and that the right platform can cut GPU bills dramatically without touching the models themselves. Strategic partnerships with Hugging Face (January 2025) and LG AI Research (July 2025, supporting LG's EXAONE 4.0 model) are early evidence of how that commercial strategy plays out in practice.

Yoo's prior playbook at Moloco scaled an AI-native company by building systems that could handle the growth that followed great technology. At FriendliAI, the play is similar but inverted: the technology is already working in production, the market is demonstrably large, and the job is to build the commercial machine fast enough to claim it before competitors do. For an operations researcher who spent a decade doing exactly that at scale, the problem is familiar. The domain is new. The math is the same.