Goldman Sachs Research: AI Buildout Will Require Hundreds of Thousands of New Energy Workers

The U.S. grid and generation workforce will need roughly 500,000 additional workers to support AI infrastructure buildout, Goldman Sachs Research concluded in an Exchanges podcast episode published April 2, anchored to a revised projection of 220% growth in global AI and data-center power demand between 2023 and 2030.

The revision, up from Goldman's earlier 175% estimate, is the figure that most concretely translates the AI buildout into near-term economic pressure. At that rate of demand growth, the bottlenecks are not primarily computational; they are physical: substation permitting timelines, transmission interconnection queues, and the scarcity of electricians, power-systems engineers, and construction labor capable of building out the required generation and grid capacity.

Brian Singer, the Goldman Sachs Research analyst who presented the findings, organized the analysis around what he called a "six P's" framework: pervasiveness, productivity, price, policy, parts, and people. The "people" category carries the most immediate implications for the firm's own workforce planning. Singer's research projects that data centers, power plants, and substations required to run large AI models will collectively drive demand for hundreds of thousands of new specialized hires in the United States alone, spanning grid engineers, site construction crews, and substation technicians.

For Goldman's investment banking and project finance teams, the research reads as a forward indicator of deal flow. Energy financing, data-center M&A, and power procurement advisory mandates are likely to grow in volume and complexity as hyperscalers race to lock in generation capacity. The "parts and policy" elements of Singer's framework add execution risk: supply-chain constraints on grid hardware, including transformers, switchgear, and cabling that remain in tight supply, and permitting bottlenecks mean projects are vulnerable to delays that extend banker workload and complicate financial modeling assumptions.

The "price" variable carries particular weight for clients and for internal Goldman AI deployments. As power demand growth outpaces grid expansion in the near term, compute pricing for data-center operators is likely to rise, compressing margins for AI product businesses and raising the hurdle rate on infrastructure investment. That dynamic matters to Goldman's own technology teams managing cloud and infrastructure spend: Singer frames AI deployments not as isolated software projects but as programs with hard dependencies on external grid timelines and construction pipelines.

For Goldman's HR and talent functions, the 500,000-worker projection is both a warning and a constraint. The firm will compete against utilities, hyperscalers, and construction contractors for the same pool of power-systems engineers and cloud infrastructure specialists, a competition with real implications for compensation benchmarks and offer timelines in those disciplines. The episode, recorded March 3 and released to the public April 2, surfaces a question that goes well beyond energy policy: at a moment when the binding constraints on AI are labor and steel, the firms that move first on infrastructure talent and energy advisory relationships will define the next cycle of deal flow before the grid catches up.