Java at the Speed of AI

Everyone's optimizing their models. Nobody's optimizing the JVM running them.

Java quietly powers the backbone of modern AI infrastructure — Spark, Kafka, Flink, and the microservices orchestrating your inference pipelines. Yet when teams chase GPU efficiency and model latency, the JVM is almost always an afterthought.

In this session, Alex Berkoff shows you what's being left on the table: how JVM tuning, smarter GC selection, and faster JIT warm-up directly translate to lower cloud spend, faster model serving, and more headroom for the workloads that actually matter. You'll also see how Azul Intelligence Cloud gives teams real-time visibility into JVM behavior across AI pipelines — so you're not flying blind when latency spikes, memory pressure builds, or an outdated ML library quietly degrades your model serving in production.


Alex Berkoff

About Alex Berkoff

Alex Berkoff is a Senior Sales Engineer at Azul, where he works with some of the world's largest Java deployments to drive cloud cost reduction and JVM performance optimization. Based in Eagle, Idaho, Alex brings over 15 years of experience as a Java software engineer and 8 years as a consultant — giving him a rare combination of deep technical fluency and the practical know-how of solving real-world problems across industries.

At Azul, Alex specializes in helping engineering teams unlock the full potential of the Java runtime — from GC tuning and warm-up elimination with ReadyNow profiles to CPU efficiency gains at scale using Azul Prime (Zing). He has worked hands-on with enterprise-scale Java workloads across Cassandra, Kafka, and microservices architectures on AWS, GCP, and OCI — partnering with companies across fintech, media, and technology to measure, prove, and operationalize JVM improvements in production.

A returning speaker on the conference circuit, Alex has previously presented on Apache Cassandra and brings that same hands-on, results-driven approach to every talk. He's passionate about making JVM performance accessible — not as theory, but as something any engineering team can measure, tune, and win with.

More About Alex »