AI infrastructure struggling to keep pace with surging demand; real economy holds bigger opportunity: Goldman Sachs

The global artificial intelligence (AI) sector is struggling to build infrastructure fast enough to keep pace with the rising demand, even as its biggest opportunity lies ahead with AI reaching the broader economy, according to a Goldman Sachs report.

At the same time, the technology industry is debating whether to slow the pace of AI development. Anthropic CEO Dario Amodei advocated slowing AI development amid concerns over the risk of losing control of increasingly advanced AI systems.

Highlighting the scale of investment needed to support growing AI demand, Goldman Sachs (GS) said, "Global hyperscaler CapEx is projected to reach over $760 billion in 2026 (approximately $2 billion per day)."


According to GS, capital investment is taking place simultaneously across geographies, with US hyperscalers accounting for the largest share of headline investments.

However, it noted, "Middle East sovereign wealth funds, European industrial investment in domestic AI capacity, and Asian capital deployment around regional supply chains are now structural components of the global buildout."

As per GS, the scale of demand is evident from Google's June 2026 agreement to pay SpaceX around USD 920 million a month, or nearly USD 30 billion through mid-2029, for access to approximately 110,000 Nvidia GPUs.

At the same time, grid operators are dealing with interconnection requests carrying energy requirements that were not anticipated in planning models three years ago, while electrical distribution equipment from major manufacturers faces multiyear backlogs.

"The mismatch arrives unevenly across sectors-first where market cycles run shortest," it stressed.

Against this backdrop, fully islanded data centres are emerging as an alternative for hyperscale AI infrastructure, particularly as grid connection delays increase as per GS.

"A year ago, fully islanded data centres were rare and not considered a mainstream solution for hyperscale AI infrastructure. Over the last 12 to 18 months, they have moved from an edge case to a credible development pathway driven by grid interconnection delays and the economics of AI compute. Estimates now suggest one-third of future capacity could be islanded, either," it said.

Beyond the infrastructure buildout, GS said software is currently showing the first major signal of AI, while the larger opportunity lies in bringing AI into the physical world and the broader economy.

"The real economy-the other ~99.5% of the global economy that AI has barely touched, from manufacturing and robotics to defense, construction, and energy- defines the actual scale of opportunity," it said.