Goldman Sachs: AI Investment Is Moving From Apps to Data Centers and Power

Goldman Sachs says AI investment is shifting from application-layer hype to core infrastructure, with capital increasingly flowing toward data centers, networking, and power systems needed to run model training and inference at scale. According to the firm's analysis, investors are moving into a 'flight to quality' phase: companies with existing infrastructure footprints are gaining attention, while narrow software plays face tougher scrutiny. Goldman Sachs Research estimates AI workloads could account for about 30% of total data-center capacity within two years. The report also projects global data-center power demand could rise roughly 175% by 2030 versus 2023, driven largely by AI compute. That puts energy supply, cooling, land access, and fiber connectivity at the center of AI economics. The implication is blunt: for the next phase of AI, power plants and substations may matter as much as model architecture. Notebook: This is the strongest mainstream signal yet that AI margins will increasingly be determined by infra ownership, not model demos.