AI inference power demand to hit 46GW by 2035

The report from ABI Research signals a fundamental shift in AI infrastructure: as models move from training to inference, power demand is expected to grow dramatically, from just 2GW in 2026 to 46GW by 2035. The rapid growth of code generation workloads, with a 52% CAGR, underscores the increasing reliance on AI for software development, a trend that will reshape data center planning and energy consumption.

The market is consolidating as major players seek to expand their capabilities. CoreWeave and Qualcomm, among others, have made acquisitions to strengthen their positions, while open-source model adoption is projected to rise from 35% in 2025 to 55% by 2026, driven by cost and flexibility concerns. These moves highlight the competitive dynamics as companies race to capture a growing market.

The report also points to challenges ahead, including the need for specialized hardware, the risk of over-reliance on AI-native startups, and the push toward edge computing to reduce costs and latency. As AI inference becomes a larger part of overall AI spending, these factors will shape the industry's evolution over the next decade.