6 in 10 Enterprises Can't Find the Root Cause When Their AI Workloads Fail | Paul Appleby, Virtana

What It Covers

Virtana CEO Paul Appleby presents findings from the AI Factory Reality Check study, revealing that 6 in 10 enterprises cannot automatically identify root causes when AI workloads fail, while governance investment lags behind infrastructure spending across banking, healthcare, airlines, and retail sectors deploying on-premises AI data centers.

Key Questions Answered

• •AI Workload Failure Rate: 6 in 10 enterprises lack automated root cause identification across AI infrastructure domains when workloads fail. Without knowing causality, remediation becomes guesswork, leaving GPU compute sitting idle and wasting capital on infrastructure that cannot deliver consistent throughput or meet production-grade reliability requirements at scale.
• •Governance Gap Risk: Enterprise AI infrastructure investment is outpacing governance and controls deployment. IT operators report increasing risk exposure while business executives push for faster AI adoption. Companies should audit whether existing governance frameworks applied to legacy infrastructure have been replicated in new AI data center environments before scaling further.
• •GPU Utilization Economics: Falling token costs do not reduce total AI spend — they accelerate token consumption as more agents deploy. Enterprises should measure GPU utilization rates alongside token costs, since throttled workloads and idle GPUs represent direct ROI loss on hardware investments that can reach hundreds of millions of dollars.
• •Observability Architecture Requirement: Effective AI factory management requires capturing telemetry across every stack layer simultaneously — data pipelines, AI orchestration, compute, network, and storage — correlating 20,000-plus metrics at sub-second intervals. Fragmented monitoring across four to six separate tools prevents real-time causality identification and blocks autonomous remediation from functioning reliably.
• •Executive-Practitioner Disconnect: Study data shows a measurable gap between c-suite AI confidence and IT operator risk assessment. Senior IT operations leaders now report infrastructure resilience metrics weekly to CEOs rather than annually. Enterprises should establish shared ROI metrics combining infrastructure cost data with business outcomes before committing to further AI scaling.