TechInsights: AI Chips to Consume Over 1.5% of Global Electricity, Generate Billions of Tons of Carbon Emissions in Next Five Years

June 28, 2024 –In a recent market outlook report, renowned research firm TechInsights has unveiled projections that underscore the continued ascendancy of data center AI chips and accelerators in the global semiconductor landscape. According to the forecast, shipments of these advanced components are anticipated to soar at a compound annual growth rate (CAGR) of 33% from 2023 to 2029, culminating in annual volumes exceeding 33 million units.

The study pinpoints generative AI applications as the primary catalyst driving this surge, with Graphics Processing Units (GPUs) emerging as the most sought-after accelerators. However, amidst this rapid expansion, TechInsights cautions that the power demands of AI technologies will constitute a significant portion of global electricity consumption, underscoring the need for balanced growth.

Illustrating this point, the report cites NVIDIA’s H100 AI chip, whose peak power consumption stands at 700 watts, comparable to that of a fully operational microwave oven. As superchips undergo iterative upgrades, their power envelopes are expected to escalate further. For instance, the NVIDIA Blackwell superchip boasts a formidable configuration comprising two 1200-watt GPUs and a 300-watt CPU, while an NVL72 rack, housing 36 such superchips, consumes a colossal 120 kilowatts.

Assuming conservative power usage of 700 watts throughout their five-year lifespan for all data center accelerators sold between 2025 and 2029, TechInsights estimates the total energy consumption of these devices to reach a staggering 2,318 terawatt-hours (TWh) over this period. Given the substantial demand for GPU capacity and the economic imperative to capitalize on these costly assets, such high utilization rates are deemed feasible.

In light of these projections, it becomes evident that AI chips will contribute approximately 1.5% to global electricity consumption over the next five years, a testament to their pivotal role in the global energy mix. This substantial energy usage, in turn, translates into significant carbon emissions. Based on the Energy Information Administration’s (EIA) data and assuming an average grid carbon intensity of 481 grams of CO2 per kilowatt-hour in 2023, TechInsights calculates that AI chips could generate as much as 11 billion tons of CO2 between 2025 and 2029, necessitating the sequestration efforts equivalent to approximately 50 billion mature trees annually.

The report underscores the dual-edged sword of AI’s relentless progress, highlighting both its transformative potential and the environmental challenges it poses. As the industry hurtles towards this AI-driven future, balancing technological advancements with sustainable energy practices will be paramount.

Tagged:

Leave a Reply