Artificial intelligence infrastructure optimization platform startup Zymtrace announced today that it has raised $12.2 million in funding, including a newly closed $8.5 million seed round, to develop its platform, expand enterprise deployments and grow its U.S. go-to-market team.

Founded in 2024, Zymtrace offers a platform designed to analyze and optimize the performance of AI workloads running on GPU-based infrastructure. The company’s approach focuses on continuous runtime profiling of both central processing unit chips and graphic processing units across a cluster to identify inefficiencies in how workloads interact with hardware resources and distributed systems.

Zymtrace’s platform works by collecting low-level execution data from production systems using an architecture based on extended Berkeley Packet Filter,  which allows instrumentation of system activity without requiring code changes. The telemetry is used to map how AI workloads move between host CPUs and attached GPUs, tracing activity down to specific code paths.

Once profiling data is collected, the platform analyzes execution patterns and generates recommendations for improving workload performance, such as adjustments to batch sizing, distributed communication patterns and kernel execution or CPU scheduling.

The platform also integrates with developer workflows and infrastructure pipelines, where automated processes can generate pull requests that implement recommended optimizations directly in the relevant code or configuration.

“Enterprises are investing heavily in AI infrastructure but operating without clear visibility into where performance is being lost,” said Zymtrace co-founder and Chief Executive Israel Ogbole. “Too often, the default response is to buy more GPUs instead of fixing the inefficiencies inside existing workloads. We’re building the visibility layer that allows AI systems to run predictably, efficiently and at scale.”

The technology is intended for organizations operating large-scale machine learning and AI inference systems in production environments, where GPU utilization, latency and throughput are closely tied to infrastructure cost. By continuously profiling production systems, the platform is used to diagnose issues that traditionally require manual investigation across multiple monitoring and debugging tools.

The $12.2 million in funding consisted of two rounds, both of which are being disclosed for the first time today, though the seed round of $8.5 million is fresh funding.

The pre-seed round of $3.7 million was led by Fly Ventures GmbH and Mango Capital, with participation from Entropy Industrial Capital. The seed round of $8.5 million was led by Venture Guides, with Mango Capital, Fly Ventures, 6 Degrees Capital and a number of individual investors also participating.

Individual investors included Hugging Face Inc. co-founder Thomas Wolf, Netlify Inc. founder Christian Bach, AI systems optimization expert Christopher Fregly and Reece Chowdhry of Concept Ventures.

“Zymtrace is creating core technology that will underpin the next generation of AI infrastructure,” said Sage Nye, partner and founding team member at Venture Guides. As infrastructure increasingly becomes the limiting factor to growth, performance gains and efficiency aren’t optional, they’re essential.”

Image: Zymtrace