Demand for AI compute capacity outpaces traditional grid-connected data center build-out, accelerating deployment of isolated data centers with dedicated power sources (renewables, small reactors).

The global artificial intelligence sector is facing significant challenges building infrastructure fast enough to meet escalating demand, according to a Goldman Sachs report. This infrastructure bottleneck is occurring even as AI’s largest economic opportunities, particularly in the physical world and broader economy, remain largely untapped.

Goldman Sachs projects that global hyperscaler capital expenditure will exceed $760 billion in 2026, equating to approximately $2 billion per day. This substantial investment is distributed across various geographies, with U.S. hyperscalers accounting for the largest share, complemented by structural capital deployment from Middle East sovereign wealth funds, European industrial investment, and Asian regional supply chains.

The scale of demand is further evidenced by Google’s agreement to pay SpaceX around $920 million monthly, totalling nearly $30 billion through mid-2029, for access to approximately 110,000 Nvidia GPUs. Simultaneously, grid operators are contending with interconnection requests that carry energy requirements far exceeding their planning models from three years ago, leading to multiyear backlogs for electrical distribution equipment from major manufacturers.

Against this backdrop, fully islanded data centers are gaining traction as an alternative for hyperscale AI infrastructure, particularly due to increasing grid connection delays. Goldman Sachs notes that a year ago, such solutions were rare, but in the last 12 to 18 months, they have moved from an edge case to a credible development pathway, driven by both interconnection delays and the economics of AI compute. Estimates now suggest that one-third of future capacity could be islanded.