AI infrastructure lags surging demand, real economy holds larger prize: Goldman Sachs
Goldman Sachs says AI infrastructure is struggling to match demand, with hyperscaler capex set to exceed $760 billion in 2026 and islanded data centres emerging as a fix.
Global artificial intelligence infrastructure is not being built fast enough to match demand, even as the sector's largest opportunity lies in extending AI into the wider economy, according to a Goldman Sachs report.
Investment scale
The bank projects global hyperscaler capital expenditure to cross $760 billion in 2026, a run rate of roughly $2 billion a day. Spending is now spread across regions rather than concentrated in one market. US hyperscalers still account for the largest share, but Middle East sovereign wealth funds, European industrial investment in domestic AI capacity, and Asian capital tied to regional supply chains have become structural parts of the global buildout.
Demand signals
The pressure is visible in the scale of individual deals. Google's June 2026 agreement to pay SpaceX about $920 million a month — close to $30 billion through mid-2029 — for access to roughly 110,000 Nvidia GPUs illustrates the level of demand.
On the supply side, grid operators are fielding interconnection requests with energy requirements that planning models did not anticipate three years ago, while electrical distribution equipment from major manufacturers faces multiyear backlogs. The mismatch surfaces unevenly, appearing first in sectors with the shortest market cycles.
Islanded data centres
With grid connections getting slower, fully islanded data centres have moved from an edge case to a credible development pathway over the past 12 to 18 months, driven by interconnection delays and the economics of AI compute. Estimates suggest about one-third of future capacity could be islanded.
Beyond infrastructure
Software is showing the first major signal of AI adoption, but the bigger prize lies in the physical world. The real economy — the roughly 99.5% of global output that AI has barely touched, spanning manufacturing, robotics, defence, construction and energy — defines the actual scale of opportunity.
Separately, the technology industry is debating the pace of development, with Anthropic CEO Dario Amodei advocating a slower approach amid concerns about losing control of increasingly advanced AI systems.