Written by: Rita

The electricity demand of global data centers is expanding at a much faster pace than expected. Goldman Sachs, in its GS SUSTAIN research report released on August 6, has significantly raised the forecast for global data center electricity demand growth by 2030 from the previous 117% to 170%, indicating that by 2030 the cumulative additional electricity consumption of data centers will exceed the total national electricity consumption of Japan in 2023. The combined capital expenditure and R&D investment from hyper-scale manufacturers are expected to surpass $1 trillion in 2026. However, Goldman Sachs also pointed out that the AI innovation cycle is shifting from the "evaluation and dream stage" to the "execution and efficiency stage," and community opposition, labor shortages, and physical environmental constraints are becoming new bottlenecks for data center construction. This article disassembles the judgment framework of this report from three dimensions: the sevenfold driving forces of electricity demand, the evolution of the innovation cycle, and the impact on sustainability.

Sevenfold Driving Forces of Electricity Demand

Goldman Sachs summarizes the factors affecting AI data center electricity demand into seven dimensions.

In terms of ubiquity, the capital expenditure expectations of hyper-scale manufacturers continue to be revised upward, with Alphabet, Amazon, and Meta recently raising their capex guidance for 2026. The combined capex and R&D of eight major hyper-scale manufacturers is expected to reach $1.2 trillion by 2026, further increasing to $1.7 trillion in 2027.

In terms of productivity, Goldman Sachs believes that the "price elasticity" of tokens and computing demand is a key variable. As long as companies are still concerned that budget cuts will affect their competitive position, or if the demand for tokens/computing is not yet fully defined, the supply-demand gap for computing power will be difficult to narrow. Goldman Sachs expects the energy efficiency improvement rate in the AI sector to continue to outpace that of the non-AI sector, but the Jevons Paradox (increased efficiency stimulating more demand) is coming into play.

In terms of pricing, higher electricity costs have not weakened the investment willingness of hyper-scale manufacturers. Goldman Sachs estimates that if the premium for green reliable electricity in the U.S. (ranging from $40 to $48 per megawatt-hour) is applied to the new electricity demand of global data centers, the total industry expenditure from 2023 to 2030 will be about $45 to $53 billion, which only accounts for 2.9% to 3.4% of the expected EBITDA of hyper-scale manufacturers in 2028.

In terms of policy, community opposition to data center construction in the U.S. is intensifying. The Governor of Texas signed an executive order on August 3, requiring data centers to undergo audits and data disclosures before receiving approval for grid access. In Virginia, there have been bipartisan calls to pause data center construction. The core concerns of community opposition focus on five aspects: electricity reliability, rising electricity prices, water resource consumption, noise pollution, and localized warming.

The supply constraints of components and manpower cannot be ignored either. Goldman Sachs estimates that by 2030 the U.S. will need an additional 300,000 jobs to meet the growth in electricity demand, with a particular shortage in technical roles like electricians. In terms of the physical environment, Goldman Sachs' analysis shows that 56% of newly established data centers globally are located in areas facing risks of high temperatures, high humidity, or drought, which will limit choices for cooling solutions and increase electricity consumption.

AI Innovation Cycle Approaching Efficiency Inflection Point

Goldman Sachs compares the AI innovation cycle to the shale gas innovation cycle, believing that AI is approaching the inflection point of shifting from the "evaluation and dream stage" to the "execution and efficiency stage." The experience from the shale gas cycle indicates that when products transition from under-demand to over-demand, financial flexibility for innovators decreases, and corporate return rates decline, the innovation cycle will enter the efficiency-driven phase.

Currently, the supply-demand pattern in the AI sector is still in a state of computing power shortage, with low vacancy rates in data centers. However, Goldman Sachs points out that three signals are worth noting. First, the reinvestment rate of hyper-scale manufacturers (capex + R&D as a share of operating cash flow + R&D) is expected to exceed 99% by 2027, indicating that financial flexibility is tightening. Second, the corporate cash return on capital employed (CROCI) is expected to decline moderately in the coming years, although it has not fallen below the lower limit of the 2014 to 2024 range, but the trend is concerning. Third, industry commentary has already shown signs of consumer-driven computing optimization.

One important lesson from the shale gas cycle is that when the industry enters the efficiency stage, thematic investment opportunities give way to stock selection. Goldman Sachs believes that if AI demand is sufficiently defined, financial flexibility tightens further, or corporate return rates significantly deteriorate, investor focus will shift from "AI thematic exposure" to "company-specific growth and differentiated returns."

AI's Sustainability Ledger: Emissions and Value Hedging

Goldman Sachs proposes an AI value assessment framework: the sustainable development benefits brought by AI minus the sustainable development costs incurred by AI. Data center emissions are the most closely monitored variable on the cost side. Goldman Sachs projects that by 2030, data center carbon dioxide emissions will increase by approximately 370 million tons (an increase of 210%) compared to 2023, equivalent to 1% of global energy emissions. If calculated using the social carbon cost of $190 per ton set by the U.S. Environmental Protection Agency, the present value of the incremental emissions in 2030 (using a discount rate of 7% to 10%) will be approximately $140 billion to $170 billion.

On the benefits side, AI applications in drug discovery have already shown significant value. Research from Goldman Sachs' biotechnology team shows that AI has increased the success rate of drug discovery by 370 basis points (equivalent to discovering 28 new drugs per year) and shortened the time from R&D to first revenue from 13 years to about 10 years. Based on a 7-year cycle, the acceleration of AI-driven drug discovery could create approximately $84 billion to $93 billion in enterprise value, which is sufficient to cover most of the social costs of incremental data center emissions.

Goldman Sachs believes that sustainable investors' participation in AI will revolve around three levels: ethics and bias prevention in the development stage, risk management and compliance, governance, and implementation. With the EU AI Act coming into effect and the U.S. SEC tightening requirements for cybersecurity disclosures, AI governance is becoming a new focus of interaction between investors and corporate management.

The surge in electricity demand from data centers is reshaping the investment logic of the entire energy supply chain. Goldman Sachs is optimistic about structural opportunities in areas such as power equipment, cooling solutions, grid infrastructure, and critical materials. The focus of the AI innovation cycle is shifting from the scale of capital expenditure to the quality of capital returns, while physical constraints on electricity supply and community permitting are impacting the pace of data center construction.

Disclaimer: This article is a compilation and interpretation by Chao Xiang Research of a third-party brokerage research report (Goldman Sachs, August 6, 2026), combined with publicly available market information. The ratings, target prices, earnings forecasts, and related judgments quoted in this article are solely the opinions of the brokerage's analysts, representing the stance of their institution, not that of Chao Xiang Research, and do not constitute any investment advice. The market carries risks; decisions must be made independently. This article should not be used as a basis for buying or selling any securities.