AI at the carbon crossroads

Solution Guides
Premium Content

This e-book outlines three critical sustainable principles that data center users of all sizes should consider when planning their HPC or AI infrastructure needs.

Data Centers AI at the carbon crossroads E-book

Exclusive Preview

AI at the Carbon Crossroads

3 Sustainable Principles for AI-Ready Infrastructure

While more traditional cloud services continue to grow fast, higher density configurations are becoming more common as GPU-driven High Performance Compute (HPC) and Machine Learning (ML) come online. The sharpest growth and highest densities of all are coming from Deep Learning and Generative AI. According to the International Energy Authority (IEA), in 2023, NVIDIA shipped 100,000 units consuming an average of 7.3 TWh of electricity per year. By 2026, the AI industry is forecasted to grow to consume at least ten times this demand. Spending in the global AI infrastructure market – including data centers, networks, and other hardware supporting AI applications – is expected to reach $423 billion by 2029, growing at a compound annual rate of 44% for the next six years.

The growth and power demands of AI put the data center industry at a carbon crossroads, where high-density power requirements have the potential to clash with emission reduction targets, and where digital transformation threatens to derail the energy transition.

Sustainability targets loom, and new regulations have been launched, making GHG reporting obligatory. These include the Corporate Sustainability Reporting Directive (CSRD), which covers over 75% of EU company turnover, as well as the EU Energy Efficiency Directive and evolving DOE and SEC requirements in the US.

Even more importantly, data center users want to report their emissions performance, with more than 23,000 companies - representing $67 trillion in market capitalization – disclosing their emissions through CDP. According to Accenture, 37% of companies have now set net zero targets.

At the same time, Rystad Energy research estimates that over $3 trillion in investment is required before 2030 to update outdated and inadequate power grids worldwide if we are to limit global warming to 1.8 degrees Celsius above pre-industrial levels. These constrained power grids cannot support the shift to AI at the same time as wholesale electrification takes place.

Bearing these factors in mind, this e-book outlines three critical sustainable principles that data center users of all sizes should consider when planning their HPC or AI infrastructure needs:
  • Redefine renewables
  • Transform operational efficiency
  • Tackle embodied impact
 

Redefine renewables

Rigorous carbon reduction measures are the key to a genuinely green digital future. AI in the current climate makes the use of low-carbon energy a critical consideration.

Most hyperscalers and a growing number of colocation providers have been growing the green grid and eliminating carbon. Data center operators are already the biggest buyers of renewables in the world, and the more sustainable colocation providers are up there with them. Iron Mountain is now one of the world’s top 20 renewable buyers. The vast majority of renewable power is purchased via Power Purchase Agreements (PPAs) which offset actual power used against renewable generation, encouraging the growth of new renewable generation projects. Results so far have been impressive, but there is much more to do.

Data Center AI at the carbon crossroads Ebook

Transform operational efficiency

Optimizing efficiency is the baseline for sustainable operations. GPU chip redesign has the potential to improve compute efficiency hugely, and data center operators need to match these efficiencies on the ground, optimizing Power Usage Effectiveness (PUE). Again, progress is being made. Denser servers with high-powered GPUs have reduced cooling requirements compared to CPUs. These days server racks that used to handle 5 kW of power can operate at 20-30 kW with the same amount of cooling due to advances in server technology. But the huge increase in overall data center GPU density raises overall power and cooling demands.

Tackle embodied impact

Data center construction comes with a high carbon price tag. Even more costly to the planet are the rare earths and metals in the servers themselves and the embodied impact of their manufacture.

Reduced carbon construction

New standards are emerging to reduce the embodied and operational impact of data center builds. These will become mainstream as the HPC and AI capacity boom gathers pace.