Zymtrace Raises $12.2M in Funding to Optimize AI Infrastructure Performance

Inside every AI company’s infrastructure bill hides an uncomfortable truth. GPUs are expensive, powerful, and wildly underused. Companies spend millions building clusters designed for speed, yet many of those machines cruise along at 35%–40% utilization. It is the technological equivalent of owning a fleet of supercars and driving them through rush hour traffic. The engines are capable of flying. The system around them keeps them crawling.

That inefficiency is exactly where Zymtrace stepped into the room. The company, founded in 2024 by Israel Ogbole and Joel Höner, just secured $12.2M across a $3.7M pre seed and an $8.5M seed round. Venture Guides led the seed with Mango Capital, Fly Ventures, and 6 Degrees Capital joining the table. The earlier round included Fly Ventures, Mango Capital, and Entropy Industrial Capital. The angel bench reads like a who’s who of people who understand infrastructure from the inside out. Thomas Wolf of Hugging Face, Christian Bach of Netlify, Christopher Fregly, Reece Chowdhry of Concept Ventures, and Jessica Thomas all leaned in.

Zymtrace is not selling hype. Zymtrace is selling sight. The platform uses eBPF powered continuous profiling to watch distributed systems without touching the application code. No restarts. No instrumentation gymnastics. Just deep visibility into how CPUs and GPUs actually behave together in production. Engineers see where performance leaks out of the stack and which lines of code are responsible. Then Zymtrace goes one step further. The system analyzes the profile and generates pull requests with suggested optimizations and projected gains. Not just observation. Action.

The technology did not appear out of thin air. Before Zymtrace existed, Israel Ogbole and Joel Höner were working inside Elastic where they helped build an eBPF continuous profiling agent that was later donated to OpenTelemetry. That agent now runs inside production environments at companies such as Cisco, Datadog, Grafana, and IBM. Zymtrace takes that lineage and pushes it deeper into the GPU era where compute is both the engine and the expense line that keeps CFOs awake at night.

The results are already showing up where it matters. In a deployment with AI company Anam, Zymtrace helped accelerate inference for the Cara3 model by 2.5x while increasing throughput by 90%. Same hardware. Same cluster. Just better understanding of how the system breathes under pressure.

Which raises a simple question sitting quietly underneath the entire AI boom. If GPUs are the gold of this era, then utilization is the refinery. And the teams that can see the inefficiencies inside the machine might end up holding something far more valuable than the hardware itself.