Offshore Oil's New Watchdog Never Sleeps

Cognite and NVIDIA deploy AI anomaly detection at Aker BP, marking a new frontier in upstream predictive maintenance

27 Mar 2026

An artificial intelligence platform designed to catch offshore equipment failures before they happen has moved from pilot program to full operational deployment, marking a significant shift in how oil and gas producers manage complex upstream infrastructure. On March 23, Cognite announced the integration of NVIDIA's NV-Tesseract AI models into its industrial data platform, with Norwegian operator Aker BP becoming the first major upstream producer to run the combined system across its offshore portfolio.

At the technical center of the deployment is a connection between Cognite's Industrial Knowledge Graph, which contextualizes operational data from thousands of offshore sensors, and NVIDIA's transformer-based time-series AI. The system continuously scans equipment data for early signs of failure across wells, heat exchangers, and other critical infrastructure. When an anomaly surfaces, an AI agent automatically assembles relevant engineering drawings and maintenance records, presenting a recommended course of action before a human engineer has begun a manual review.

The need for such automation is considerable. Aker BP collects roughly one million data streams across its offshore operations, yet historically only a fraction have been monitored in real time. Engineers have typically relied on manual troubleshooting with basic diagnostic tools, a workflow that the new platform substantially compresses. Paula Doyle, Aker BP's chief digital officer, said the aim was not smarter alarms but engineers freed from chasing false leads. Chirayu Shah, Cognite's chief product officer, said high-quality, contextually rich operational data remains the central challenge for AI adoption across the sector.

The announcement was made at CERAWeek in Houston, lending the deployment additional visibility among producers who face growing pressure to extract more efficiency from aging infrastructure. Predictive maintenance now accounts for close to 38 percent of all upstream AI spending globally, according to industry estimates, and the broader oil and gas AI market is forecast to reach $7.91 billion by 2031.

Whether the Aker BP model translates cleanly to other operating environments, including the large, complex portfolios common across the Middle East, remains an open question. The architecture is designed to be adaptable, but field conditions, data quality, and institutional readiness vary considerably across producers. How widely the approach scales will likely determine its influence on the next generation of upstream operations.