WEKA's AI Expert Warns Industry to Prepare for Regulatory Requirements

Val Bercovici, Chief AI Officer at WEKA, is an expert in developing and advising emerging technologies for supporting increasingly data-intensive AI systems. With a career spanning from shaping NetApp's early cloud strategy to building AI infrastructure at WEKA, he has seen the industry evolve. The bottleneck moves, and the industry takes years to notice, Bercovici explained.

Bercovici joined WEKA in January 2025 as Chief AI Officer, focusing on advancing the technologies that underpin next-generation artificial intelligence. He concentrated on building AI agent infrastructure, accelerating training and inference workloads, and improving the economics of AI compute. The company's software-defined platform is designed for demanding data requirements of AI, machine learning, high-performance computing, and other accelerated workloads.

The industry has moved from compute becoming elastic to orchestration being the new choke point. Bercovici realized that WEKA's technology could serve memory applications with microsecond-level latency, a lightbulb moment for him. This technology can help eliminate storage bottlenecks, improve GPU utilization, and accelerate AI model training and inference.

Bercovici emphasized the importance of preparing infrastructure for regulatory requirements related to AI safety frameworks. Companies shouldn't wait for the final checklist, but instead focus on demonstrating what their AI models did, what data they touched, and how they behaved at a specific point in time. The EU AI Act became enforceable this month, classifying most agent orchestration as high-risk.

The solution to building safe AI is more AI, applied optimally and very intentionally, according to Bercovici. He predicts that safety testing will create new infrastructure demands, with workloads looking different from conventional model training or inference. Safety training and testing are bursty, read-heavy, and comparative, generating and consuming enormous amounts of intermediate state.