Goldman Sachs: Data Centres Set for 165% Electricity Spike

Global energy markets are undergoing a profound transformation driven by the AI boom, with data centres projected to consume 165% more electricity by 2030 compared to 2023, according to new Goldman Sachs research.

Goldman Sachs reports US data centre construction spending has tripled in the last three years as tech leaders rapidly scale up infrastructure for cutting-edge AI systems. 

Despite frequent new facility launches, third-party occupancies are nearing historical highs throughout most US markets, highlighting persistently strong demand.

James Schneider, Goldman Sachs’s Senior Equity Research Analyst covering digital infrastructure, says: “Over the next five to six years, we forecast substantial demand growth in the global data centre market.”

The magnitude of the ongoing shift emerges when reviewing present-day power usage trends.

Right now, global data centres consume about 55GW of electricity, with more than half dedicated to cloud computing workloads.

Traditional business applications, such as email and file storage, make up roughly a third, while AI workloads represent just 14% of total consumption

By 2027, Goldman Sachs anticipates total demand soaring to 84GW.

How the power density revolution is changing the infrastructure game

Goldman Sachs believes the composition will change dramatically, with AI claiming more than a quarter of all power usage while cloud computing’s share drops to half and traditional workloads shrink to less than a quarter.

Where a conventional data centre server rack would draw a certain level of power, its AI counterpart uses up to ten times more electricity.

This demand is only set to intensify, with power density in facilities projected to climb from 162kW per square foot today to 176kW by 2027.

“Data centre supply – specifically the rate at which incremental supply is built – has been constrained over the past 18 months,” James says, due to both the heightened power needs and the growing density of contemporary AI infrastructure.

Put simply, a single query to ChatGPT, OpenAI’s conversational AI, consumes 2.9 watt-hours of electricity, almost 10 times the energy required for a Google search, according to data from the International Energy Agency (IEA).

When scaled across millions of interactions daily, the energy impact becomes considerable.

Goldman Sachs outlines three potential growth scenarios through 2027.

The base forecast anticipates demand rising 50% to 92GW. If AI adoption slows, growth may ease to 14%.

But if demand surpasses estimates, driven by more power-intensive chips or faster AI integration, annual growth could accelerate to 20%.

This trend echoes previous tech cycles, although the pace and magnitude feel unprecedented.

From 2015 to 2019, data centre workloads nearly tripled while annual power consumption remained steady at around 200TWh.

Efficiency advancements compensated for the workload surge.

Since 2020, however, those efficiency gains have decelerated just as AI demand has sharply increased, fuelling the current upswing.

European market facing 170GW pipeline: explained

Goldman Sachs projects that Europe is facing a data centre pipeline of approximately 170GW, equivalent to about one-third of the continent’s total current electricity consumption.

Alberto Gandolfi, Managing Director, Equity Research at Goldman Sachs, says: “Inflecting power demand is monumentally important, because it’s been declining for 15 years in Europe.”

By 2030, European data centres alone are expected to require as much electricity as Portugal, Greece and the Netherlands consume combined today.

The geographical spread of this growth follows familiar trends.

Nations with inexpensive, plentiful renewable energy – particularly the Nordic nations, Spain and France – are attracting investment due to their power advantages.

Meanwhile, financial and technology centres such as Germany, Britain and Ireland compete through tax incentives and well-established business ecosystems.

In the Asia Pacific region, major hubs like Beijing and Shanghai continue their expansion.

Although this region has added the most data centre capacity in the past decade, North America now leads in planned new developments.

The market itself is increasingly consolidating around hyperscale operators – the select few companies that provide cloud services on a massive scale.

Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform are the dominant players, with other significant cloud providers also included in this group.

Goldman Sachs expects that, by 2030, these hyperscalers and wholesale operators will control 70% of global capacity, up from 60% today.

Why grid strain demands US$720bn investment

Meeting the rising demand will necessitate massive infrastructure investments.In the US alone, utilities must invest US$50bn in new generation capacity dedicated to data centres, while global grid upgrades could total US$720bn by 2030.

“Retrofitting existing facilities to support these massive jumps in power density is becoming complex and compromised. We will need new, purpose-built AI infrastructure to power the next generation,” says Frank Long, a Vice President at the Goldman Sachs Global Institute, where he focuses on AI.

The challenge is heightened by the fact that efficiency gains, which previously offset rising power demands, have largely plateaued.

Data centre occupancy rates also reflect the supply-demand imbalance: from 85% in 2023, occupancy is expected to peak above 95% by late 2026 before easing as new sites come online.

Such high capacity utilisation generally indicates market stress and upward pressure on prices.

The energy mix fuelling this growth is shifting towards renewables, though baseload power remains crucial.

Goldman Sachs projects that 40% of new capacity will be sourced from renewable energy.

The economics are favourable, with onshore wind costs approximately US$25 per megawatt-hour and solar about US$26, compared with US$37 for combined cycle natural gas before factoring in carbon costs.

However, renewables face operational limitations: utility-scale solar runs around six hours daily on average, and wind facilities operate about nine hours.

Since data centres require continuous power, operators are leaning toward hybrid setups that blend renewables with battery storage and backup natural gas capacity.

Nuclear power gaining support from the technology sector

The reliability demands of AI infrastructure are reviving interest in nuclear power. 

Over the past year, US technology firms have signed contracts for more than 10GW of potential new nuclear capacity, with Goldman Sachs highlighting three plants that could be operational by 2030.

The argument for nuclear power centres on its ability to deliver steady baseload electricity without carbon emissions.

Large-scale onsite nuclear generation costs roughly US$77 per megawatt-hour when carbon pricing hits US$100 per tonne, making it competitive with gas-fired options once environmental costs are factored in.

Brian Singer from Goldman Sachs Research says how technology companies’ sustainability commitments align with their reliability needs, creating momentum for both refurbishing decommissioned plants and building new reactors. 

The political landscape has shifted as well, with nuclear power gaining bipartisan support in the US and countries like Switzerland reconsidering their nuclear phase-out plans.

At the same time, recent developments introduce uncertainty into these outlooks.

DeepSeek, a Chinese AI model reportedly on par with leading US systems but requiring fewer computational resources, raises questions about future infrastructure demands.

If DeepSeek’s efficiency gains prove scalable, they could temper the expected surge in power consumption.

James acknowledges the uncertainty around DeepSeek’s “training, infrastructure and ability to scale” while maintaining Goldman Sachs’s broader forecasts. 

The firm also notes that cooling systems consume 35-40% of hyperscalers’ energy use regardless of the AI technology involved.

“Longer term, we see potential for a significant reduction of data centre emissions intensity and potentially in absolute emissions, as more nuclear power comes online and AI computing shifts to using AI models as opposed to training them,” the analysts say.