AI might soon use more electricity than bitcoin mining and some countries

AI's energy use may exceed Bitcoin's by 2025, disrupting global climate goals.

: AI is on track to demand more electricity than Bitcoin mining by late 2025, potentially impacting power grids and climate commitments. The growth of AI has led to dramatic increases in data center construction and energy consumption. Research by Alex de Vries-Gao highlights potential energy consumption that could eclipse Ireland's national energy use. This surge is driven by increased production of specialized AI hardware and the infrastructure needed to support growing AI applications.

Advancements in AI technology, particularly in generative AI, have fueled a massive increase in data centers and the hardware necessary to support complex AI models. According to Alex de Vries-Gao from Vrije Universiteit Amsterdam, AI's electricity consumption is expected to surpass that of Bitcoin mining before the end of 2025. This shift signifies a significant challenge for global power grids and could have far-reaching environmental implications. His research, published in Joule, utilizes public device specs, estimates from analysts, and corporate disclosures to form these predictions.

The energy demand from AI is largely due to the proliferation of AI accelerators from companies like Nvidia and AMD. These machines are designed specifically to handle the intensive tasks that AI models require, such as training and inference for large datasets. The energy consumption is projected to rise to account for nearly half of all data center electricity usage by 2026, a marked increase from around 20 percent currently. This transformation underscores the urgent need for sustainable strategies in handling this shift.

One of the cores of this burgeoning energy demand is the CoWoS packaging technology by TSMC, which allows for the integration of powerful processors and high-speed memory into single units. Despite TSMC's efforts to expand its production, which saw capacity more than double between 2023 and 2024, demand from AI chip manufacturing giants like Nvidia continues to exceed supply. The projected AI power needs could reach 23 gigawatts by the end of 2025, nearing the UK's average national consumption rate.

As companies like Google face what they describe as "power capacity crises," there have been shifts to repurpose fossil fuel infrastructure, including securing 4.5 gigawatts of natural gas capacity for AI workloads. This indicates a potential environmental backlash, especially if the expansions occur in regions heavily reliant on fossil fuels. The disparity in carbon emissions between regions powered by different energy sources highlights the importance of location in determining the overall ecological impact of AI operations.

Transparency remains a critical issue in this expanding industry, with tech companies often not disclosing the locations or methods of their energy use for AI technologies. This opacity makes it difficult for policymakers and climate researchers to accurately gauge or prepare for the environmental challenges posed by the AI sector's rapid expansion.

Sources: TechSpot, Cell.com

AIdata centerEnergy