Why Your GPU Investments Are Running at 40% Efficiency, and How Zymtrace Fixes It

What if your multi-million-dollar GPU investment was quietly delivering less than half its potential? In this episode of Built Not Born, the conversation focuses on why AI infrastructure performance often lags far behind spend, and how continuous profiling is emerging as a real competitive edge. Join host Sage Nye as she sits down with Israel Ogbole, CEO and cofounder of zymtrace, to explore why most GPU clusters run at just 20–40% utilisation, how traditional observability tools miss the real bottlenecks, and why profile-guided optimisation can unlock massive efficiency gains without demanding deep CUDA expertise. Israel shares hard-won lessons from building a must-have infrastructure product: validating real willingness to pay, avoiding “nice-to-have” traps, and navigating early founder dynamics. The discussion also dives into the growing talent gap in low-level systems engineering, and how rich profiling data can become context for LLM-driven optimisation workflows. It’s a practical, insight-dense conversation for founders, operators, and AI leaders focused on extracting real returns from AI infrastructure, not just scaling hardware.