Mint Explainer: Why the surge in use of AI and data farms could pose challenges for India's power sector

The peak demand exceeded revised forecasts by government power sector experts. The Central Electricity Authority (CEA) had forecast peak demand at 235GW in May and over 250GW in August and September.

The electricity regulator, in its power survey released in November 2022, had projected peak demand at around 245 GW in 2024-25 and 260 GW in 2025-06. This demand is projected to increase to 328 GW in 2031-32 and 488 GW in 2041-42.

CEA's electricity demand forecast is based on demand for residential, industrial, commercial, irrigation, rail traction, public water and public lighting.

Additionally, the 2022 Power Study took into account the emerging demand for electric vehicles, assuming that 30% of all vehicles sold in 2030 will be battery electric vehicles, rising to 100% by 2042. The contribution of solar photovoltaics and hydrogen missions, which will ease pressure on traditional energy sources, was also taken into account.

However, the widespread adoption of artificial intelligence (AI), particularly generative AI, and the creation of large-scale data centers to support online AI interactions is likely to exceed these demand projections. mint Learn why the rise of AI could lead to a surge in demand for electricity.

Why should utilities be concerned about the increased use of AI?

Concerns about AI's energy impact are growing around the world. In its annual report on electricity market trends and policies released in January, the International Energy Agency estimated that data centers, cryptocurrencies, and AI will consume a combined 460 terawatt-hours (TWh) of electricity in 2022, accounting for about 2% of global electricity demand.

This demand could grow to 620-1,050 TWh by 2026. The agency's base case for demand is just over 800 TWh. The additional demand from data centers, cryptocurrencies and AI by 2026 is equivalent to Sweden's electricity consumption in 2022 at the low end and Germany's at the high end.

According to an article published by the World Economic Forum, the energy requirements to run AI tasks are already accelerating at an annual growth rate of 26-36 percent. “This means that by 2028, AI could use more electricity than the entire country of Iceland used in 2021,” the article reads.

A widely cited, peer-reviewed analysis by Alex de Vries, a doctoral student at Vrije Universiteit Amsterdam, highlights that AI servers are power-hungry devices: “A single NVIDIA DGX A100 server can consume as much electricity as several US homes combined, so the power consumption of hundreds of thousands of these devices would quickly add up,” de Vries wrote in his blog.

His paper was published in a scientific journal. Joule October 2023. AI systems require graphics processing units (GPUs) to perform large amounts of calculations quickly, 95% of which are provided by Nvidia.

He predicts that global AI-related electricity consumption could increase by 85.4 to 134.0 TWh by 2027. This represents 0.5% of global electricity consumption and is comparable to the annual electricity consumption of countries such as the Netherlands, Argentina, and Sweden.

Why is AI so power-hungry?

according to New ScientistIn his paper, de Ville noted that if Google were to switch its entire search business to AI, it would use 29.3 terawatt-hours of electricity per year — equivalent to the electricity consumption of Ireland and nearly double the company's total energy consumption of 15.4 terawatt-hours in 2020.

“Such a switch would be virtually impossible because it would require more than four million powerful computer chips called GPUs, which are currently in high demand but in limited supply. That would cost $100 billion, a figure that even Google's deep pockets would have trouble financing,” he wrote.

The IEA estimates that a single Google search consumes 0.3 watt-hours of power, while a ChatGPT request consumes 2.9 watt-hours of power.

AI-enabled online interactions, such as chatbots, require the support of massive data centers with hundreds of machines equipped with powerful chips. Not only do these data centers need to run 24/7, but they also need energy to cool the machines – and this is just a fraction of the energy required for AI.

The training and inference phases of an AI model also consume a lot of power. During the training phase, the model processes vast amounts of data to learn and evolve. During the inference phase, the model goes to work and answers specific real-world problems or queries. Both phases consume a lot of power, with the inference phase accounting for the majority of the consumption.

By one estimate, training a large language model like OpenAI’s GPT-3 consumes around 1,300 MWh of electricity, equivalent to the annual consumption of around 130 US households.

Will India be able to meet the surge in electricity demand that will come with widespread adoption of AI?

India is rapidly digitizing, which will itself lead to increased demand for electricity, and a joint Microsoft and LinkedIn study found that more than 90% of Indian knowledge workers are using AI tools to improve productivity.

The increasing use of generative AI models such as ChatGPT by students, professionals, and industries will lead to the establishment of many large data centers.

India is already one of the world's leading data center locations and as data localization increases, the number of data centers will increase. These data centers require a reliable and stable power supply.

Governments, regulators, and power generation and distribution companies must take this demand into account when forecasting to ensure sufficient capacity is available.