TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
Goldman Sachs Research Report Analysis: Global Data Center Power Demand to Surge 170%, AI Innovation Cycle Approaching Efficiency Inflection Point

Goldman Sachs Research Report Analysis: Global Data Center Power Demand to Surge 170%, AI Innovation Cycle Approaching Efficiency Inflection Point

2026.08.07
Share
This article deconstructs the judgment framework of this report from three dimensions: the seven drivers of electricity demand, the evolution of innovation cycles, and sustainability impact.

By: Rita

Global data center power demand is expanding at a pace far exceeding expectations. In the GS SUSTAIN report released on August 6, Goldman Sachs significantly raised its forecast for the growth rate of global data center power demand by 2030 from the previous 117% to 170%, implying that the cumulative incremental power consumption of data centers by 2030 will exceed Japan's total national power consumption in 2023. The combined capital expenditure and R&D investment of hyperscalers are expected to exceed $1 trillion in 2026. However, Goldman Sachs also pointed out that the AI innovation cycle is shifting from the "assessment and dream phase" to the "execution and efficiency phase," with community opposition, labor shortages, and physical environmental constraints becoming new bottlenecks restricting data center construction. This article deconstructs the judgment framework of this report from three dimensions: the seven drivers of power demand, the evolution of the innovation cycle, and sustainability impacts.

Seven Drivers of Power Demand

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

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

In terms of productivity, Goldman Sachs believes the "price elasticity" of token and compute demand is a key variable. As long as companies remain concerned that cutting budgets will affect their competitive position, or if the demand for tokens/compute has not been fully defined, the supply-demand gap in compute power will be difficult to narrow. Goldman Sachs expects the pace of energy efficiency improvement in the AI field to continue to be faster than in non-AI fields, but the Jevons paradox (efficiency improvements stimulating more demand) is at play.

In terms of price, higher electricity costs have not weakened the investment willingness of hyperscalers. Goldman Sachs estimates that if the premium for green reliable power in the US ($40 to $48 per megawatt-hour) is applied to the global data center new power demand, the total industry expenditure from 2023 to 2030 will be approximately $45 to $53 billion, accounting for only 2.9% to 3.4% of the hyperscalers' expected EBITDA in 2028.

In terms of policy, community opposition to data center construction in the US is intensifying. On August 3, the Governor of Texas signed an executive order requiring data centers to undergo audits and data disclosure before obtaining grid access approval. In Virginia, there have been bipartisan calls to pause data center construction. The core concerns of community opposition focus on five aspects: power reliability, electricity price hikes, water consumption, noise pollution, and local temperature rise.

Supply constraints on components and labor are also cannot be ignored. Goldman Sachs expects that by 2030, the US will need to add approximately 300,000 jobs to meet power demand growth, with shortages of skilled trades such as electricians being particularly prominent. In terms of the physical environment, Goldman Sachs analysis shows that 56% of new data centers globally are located in areas facing risks of high temperature, high humidity, or drought, which will limit the selection of cooling solutions and drive up power consumption.

AI Innovation Cycle Approaching Efficiency Inflection Point

Goldman Sachs draws an analogy between the AI innovation cycle and the shale gas innovation cycle, believing that AI is approaching the inflection point of shifting from the "assessment and dream phase" to the "execution and efficiency phase." Experience from the shale gas cycle indicates that when products shift from supply shortage to oversupply, innovators' financial flexibility declines, and corporate returns deteriorate, the innovation cycle enters the efficiency-driven phase.

The current supply-demand landscape in the AI field remains in a state of compute shortage, with data center vacancy rates remaining low. However, Goldman Sachs points out three signals worth watching. First, the reinvestment rate of hyperscalers (proportion of capex + R&D to operating cash flow + R&D) is expected to exceed 99% in 2027, and financial flexibility is narrowing. Second, Cash Return on Capital Invested (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, the trend is worth vigilance. Third, industry commentary has shown signs of consumers driving compute optimization.

An important lesson from the shale gas cycle is that when the industry enters the efficiency phase, opportunities for thematic investment give way to stock selection. Goldman Sachs believes that if AI demand is fully defined, financial flexibility tightens further, or corporate returns deteriorate significantly, investors' focus will shift from "AI theme exposure" to "company-specific growth and differentiated returns."

AI Sustainability Ledger: Hedging Emissions and Value

Goldman Sachs proposes an AI value assessment framework: sustainable development benefits brought by AI minus sustainable development costs brought by AI. Data center emissions are the most concerned variable on the cost side. Goldman Sachs expects that by 2030, data center carbon dioxide emissions will increase by approximately 370 million tons compared to 2023 (a 210% increase), equivalent to 1% of global energy emissions. Calculated based on the US EPA's social cost of carbon of $190 per ton, the present value of incremental emissions in 2030 (7% to 10% discount rate) will be approximately $140 to $170 billion.

On the benefit side, the application of AI in the field of drug discovery has shown considerable value. Research by Goldman Sachs' biotechnology team shows that AI has increased the drug discovery success rate by 370 basis points (equivalent to discovering 28 more new drugs annually) and has shortened the time from R&D to first revenue from 13 years to about 10 years. Calculated on a 7-year cycle, AI-driven drug discovery acceleration can create approximately $84 to $93 billion in corporate present 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 and control in the development stage, risk management and compliance, and governance and implementation. With the EU AI Act taking effect and the US SEC's requirements for cybersecurity disclosure becoming stricter, AI governance is becoming a new focus of interaction between investors and corporate management.

The surge in data center power demand is reshaping the investment logic of the entire energy industry chain. Goldman Sachs is bullish on structural opportunities in fields such as power equipment, cooling solutions, grid infrastructure, and key materials. The focus of the AI innovation cycle is shifting from the scale of capital expenditure to the quality of capital returns, with physical constraints on power supply and community licensing operations affecting the pace of data center construction.

Disclaimer: This article is a compilation and interpretation by TechFlow Research of a third-party broker research report (Goldman Sachs, August 6, 2026), combined with organized public market information. The ratings, target prices, earnings forecasts, and related judgments cited in the article are the views of the broker's analysts, represent only the position of their affiliated institution, do not represent the views of TechFlow Research, and do not constitute any investment advice. The market involves risks; decisions must be made independently. This article should not be used as a basis for buying or selling any securities.