High energy consumption of AI, GenAI raises power shortage fears

The exponential growth in consumption has raised concerns about power availability and its impact on sustainability goals.

Gartner anticipates that 40% of current AI data centres will face operational constraints due to power availability by 2027. 

“The explosive growth of new hyperscale data centres aimed at implementing GenAI is creating an insatiable demand for power that will surpass utility providers' capacity to expand quickly enough,” said Bob Johnson, VP Analyst at Gartner. “This threatens to disrupt energy availability, leading to shortages that will hinder the development of new data centres for GenAI and other applications from 2026 onwards.” 

Meanwhile, Gartner estimates that the power required to operate incremental AI-optimised servers will reach 500 terawatt-hours (TWh) annually by 2027, around 2.6 times the demand in 2023. 

“Larger data centres are being planned to manage the vast amounts of data necessary for training and implementing the rapidly advancing large language models (LLMs) that support GenAI applications. However, short-term power shortages are likely to persist for years, as establishing new power transmission, distribution, and generation capacity is time-consuming and will not resolve current issues,” Johnson noted.

In the near future, the establishment of new data centres and the growth of GenAI will hinge on the availability of power. Gartner advises organisations to assess the risks posed by potential power shortages on all products and services.

The unavoidable outcome of the anticipated power shortages will be a rise in electricity prices, which will further escalate the operational costs of LLMs, according to Gartner. 

Major power users are already negotiating long-term contracts to secure independent power sources. 

"Data centre operators will have to pay a premium to secure reliable power," said Johnson. "These costs will be passed on to AI/GenAI product and service providers, ultimately affecting consumers." 

Gartner recommends organisations factor in higher power costs when developing new AI products and services, while exploring alternative approaches requiring less power.

“Significant power consumers are collaborating with major suppliers to secure long-term, guaranteed power sources independent of other grid demands,” Johnson explained. “In the interim, the cost of electricity to operate data centres will substantially rise, as operators leverage economic pressures to secure necessary power. These increased costs will inevitably be passed on to AI/GenAI product and service providers.” 

Gartner encourages organisations to review future plans with an eye on increased electricity costs and to negotiate long-term contracts for data centre services at reasonable rates. 

Furthermore, organisations should account for potential cost spikes when devising plans for new products and services, while also exploring alternative methods that require less power.
 

Sustainability goals are also likely to be adversely impacted by short-term measures to enhance power supply, as surging demand is compelling suppliers to boost production by any means available. In some instances, this entails keeping fossil fuel plants that were scheduled for decommissioning operational beyond their intended shutdown dates. 

“The reality is that greater use of data centres will lead to increased CO2 emissions to generate the necessary power in the short term.  This will complicate efforts for data centre operators and their clients to meet ambitious sustainability targets related to CO2 emissions,” Johnson stated. 

While renewable energy sources like wind and solar are attractive, they cannot provide the 24/7 power availability data centres require. Therefore, alternative supply options during non-generating periods are needed. 

According to Gartner, only hydroelectric, fossil fuel, or nuclear power plants can reliably provide 24/7 power. Fortunately, new technologies like advanced battery storage (eg, sodium-ion batteries) and small nuclear reactors are emerging to provide cleaner and more sustainable power sources in the long run.

Gartner recommends that organisations revisit their sustainability targets regarding CO2 emissions given upcoming data centre demands and power sources over the next few years. When developing GenAI applications, they should prioritise minimising computational power usage and consider the potential of alternatives such as edge computing and smaller language models.

By taking proactive measures, organisations can navigate the challenges of surging AI power demands and achieve their AI goals while safeguarding their sustainability commitments.