AI Expansion to Trigger a 220 Percent Surge in Global Energy Demand, Goldman Sachs Warns

2026-04-04 economy

New York, Saturday, 4 April 2026.
Goldman Sachs warns that rapid artificial intelligence expansion will spike global power consumption by 220 percent by 2030, an increase equivalent to adding a major country to the grid.

The Hyperscale Shift: Rewriting the Energy Equation

In a newly revised forecast released in early April 2026, Goldman Sachs significantly elevated its projections for global power consumption tied to artificial intelligence and data centers [1]. The investment bank now anticipates a staggering 220 percent growth in global AI and non-AI data center power demand between 2023 and 2030, marking a 25.714 percent increase in their own growth forecast compared to their previous 175 percent estimate [1][6]. Brian Singer, head of GS SUSTAIN, noted in a March 2026 podcast that this absolute surge is comparable to integrating the world’s sixth-largest power-consuming nation directly into the global grid [1][2][6]. The primary catalyst for this revision is the unexpectedly high energy intensity of “inference”—the phase where trained AI models generate predictions and content based on new data [1].

The architectural demands of modern AI are fundamentally rendering legacy infrastructure obsolete [GPT]. Traditional data center racks historically operated at capacities of 5 to 10 kilowatts (kW), but modern AI racks are now routinely exceeding 50 to 100 kW [8]. Because air cooling becomes thermodynamically unviable beyond the 30 to 40 kW threshold, hyperscale facilities are being forced to adopt advanced liquid cooling technologies [8]. This escalating compute density means that power availability is rapidly transitioning from a mere operational expense to a critical strategic constraint [6][8]. In high-growth technological corridors such as Bengaluru and Hyderabad, reliable baseload electricity is now viewed as the ultimate asset constraint, directly impacting real estate valuations and capital expenditure cycles [6].

Infrastructure Bottlenecks and the Decarbonization Dilemma

The physical buildout required to support this 220 percent surge is struggling to keep pace with the technology sector’s demands [1][8]. Traditional data center construction timelines, which typically span 24 to 36 months, are fundamentally misaligned with the rapid deployment cycles of AI [8]. Furthermore, the sheer volume of electricity required places hyperscaler expansion on a direct collision course with global decarbonization targets [6]. As utility companies and tech firms rush to secure energy, they face a complex trilemma: balancing grid stability, ensuring energy security, and adhering to strict carbon reduction mandates [6]. Consequently, infrastructure reliability has emerged as a dominant investment theme for 2026, though the feasibility of powering these multi-megawatt deployments solely with green energy remains questionable [alert! ‘The exact cost and timeline for deploying sufficient green energy infrastructure to meet this specific AI demand spike remains highly uncertain and variable by region’] [6][8].

Labor Market Paradox: Job Creation Versus Displacement

The macroeconomic ripple effects of this AI boom are acutely visible in the United States labor market, which is experiencing a profound structural realignment [3]. To support the physical expansion of data centers, Goldman Sachs projects the U.S. will need to generate approximately 500,000 new jobs by 2030 [1][3]. This includes 300,000 roles dedicated to building new power generation facilities and an additional 200,000 jobs focused on upgrading transmission and distribution grids [1]. However, this infrastructure-led job creation stands in stark contrast to the broader automation risks. Globally, an estimated 300 million jobs are exposed to AI automation, and Goldman Sachs projects that 6 to 7 percent of U.S. workers could be entirely displaced over the next decade as AI automates tasks that currently account for 25 percent of all U.S. work hours [3].

This dichotomy is already influencing monetary policy considerations in the spring of 2026 [3]. The U.S. labor market has recently exhibited signs of slowing, with the national unemployment rate ticking up from 4.3 percent to 4.5 percent [3]. Joseph Briggs, co-leader of the global economics team at Goldman Sachs, highlighted that if AI-driven job losses are accelerated—or “pulled forward”—it could result in economic underperformance relative to baseline forecasts [3]. Such a scenario might compel the Federal Reserve to initiate interest rate cuts to stabilize the employment mandate [3].

The “Productivity Marathon” and Equity Markets

Despite the infrastructural and labor challenges, the financial sector remains highly optimistic about AI’s capacity to drive corporate profitability [5][7]. Following a sharp 10 percent global market correction in March 2026—a drawdown that Goldman Sachs’ Chief U.S. Equity Strategist Ben Snider described as having “cleared the decks” of speculative froth—the firm issued a highly bullish mid-year outlook on April 3, 2026 [5][7]. Goldman Sachs established a year-end price target of 7,600 for the S&P 500, which implies a 12 percent total return driven by robust earnings growth [5][7]. The market is effectively transitioning from an infrastructure-heavy focus to an “Execution” phase, dubbed a broader “productivity marathon” [5][7].

This productivity narrative is moving beyond theoretical projections into tangible corporate workflows [5]. Goldman Sachs itself serves as a prime case study; as of April 1, 2026, the bank’s Chief Information Officer Marco Argenti revealed that the firm has deployed an internal AI assistant to 47,000 employees [4]. This tool logs over a million prompts monthly, drastically reducing software build times from months to days and allowing the bank to terminate several third-party vendor contracts [4]. Across the broader market, Goldman Sachs anticipates that AI-driven productivity will contribute a direct 0.4 percent uplift to S&P 500 earnings per share (EPS) in 2026, a figure expected to nearly triple by 2027, providing a fundamental floor for current market valuations [5][7].

Sources

Artificial intelligence Energy infrastructure