Data centers are at a crossroads between AI and sustainability

Data centers are growing exponentially, as are their energy needs. Legacy data centers were not designed for today’s use cases, including and especially the unique and fast-growing computing demands of AI.

And it’s not just the data centers themselves. The electrical grid was designed a century ago and, even as it adopts new energy generation technologies, it is struggling to serve the concentrated increase in energy demands driven by unprecedented levels of computing requirements. At the same time as their computing demands are rising, companies need to find ways to reduce their carbon emissions, a large portion of which come from their data centers, in order to meet their sustainability commitments.

Better data centers — ones built for 2030, not 2020 — can help.

Sidewalk Infrastructure Partners (SIP) is focused on the intersection of technology and infrastructure, taking innovative and creative approaches to make the future of infrastructure sustainable, resilient, and inclusive. From our vantage point, we believe the data center is ripe for innovation — and disruption.

To dive into the topic of how technology and innovation can help solve some of the existential challenges currently facing the data center industry, SIP recently convened leaders from across the space for a Future of Data Centers Summit, which was sponsored by Google Cloud. The summit featured a keynote from Dr. Arun Majumdar, the Dean of the Stanford Doerr School of Sustainability and Founding Director of Advanced Research Projects - Energy, and panels on data center sustainability, electrical grids, and the impact of AI, with a range of distinguished speakers from the private and public sectors. Leaders from academia, industry, and government discussed how data centers must evolve to meet these challenges. Their commentary and insights demonstrated both the importance of these issues and the forward-looking leadership that can help address them.

Dr. Arun Majumdar's keynote presentation. keepAspectRatio

One of the core challenges that the industry faces is data centers’ relationships with the electrical grid. The fundamental architecture of the electrical grid has remained largely unchanged for over a century, despite significant technological advancements and higher-than-ever demands. This system relies on a diverse range of generation sources, all of which must be matched in real-time to serve the needs of energy consumers. The proliferation of distributed energy resources, especially electric vehicles, will continue to accelerate, causing additional strain.

Data centers add considerable stress to our already-vulnerable electrical grid. As demand for data processing and storage surges, power consumption in these facilities skyrockets to maintain 100% availability for highly variable load. Unfortunately, the strain on the electrical grid is expected to increase dramatically. The availability of sufficient electricity at the local level has become a constraint to siting data centers in some of the world’s fastest growing markets, from Dublin, Ireland, to Loudoun County, Virginia. Notably, while data centers are estimated to consume 1-3% of global electricity, in specific locations they can consume significantly more: in Ireland, for instance, they are estimated to consume 14% of the nation’s electricity.

Compounding this problem is the rise of AI. As demand for AI, including generative AI tools like Bard and ChatGPT, increases dramatically, so do AI-driven computing power and needs. Importantly, some AI compute is also fundamentally different than traditional compute, in that it is batchable and interruptible. This presents both a challenge and an opportunity for data centers: each inquiry to an AI system requires vast amounts of power, but the power for all such computation does not necessarily have to be supplied continuously, unlike traditional compute.

There is also a fundamental tension between the speed at which AI-related capacity is developing and the need for lengthy lead-times in building data centers or transmission and distribution upgrades to the grid. The outlook for AI and its corresponding compute requirements looks different today than just six months ago, making long-term forecasts particularly challenging. At the same time, customers need data centers that process both traditional and AI-driven compute, so flexibility and adaptability should be central to new data center designs.

These two trends — electrical grid challenges, and the surge of data center demand driven by AI – arrive at a paradoxical inflection point, as most countries and global corporations align on acting to address climate change. Data centers, as significant energy consumers, have a crucial role to play in helping corporations achieve these objectives. Achieving real sustainability in data centers will be difficult, but not impossible. Truly carbon-neutral or carbon-negative data centers will require a complete redesign of today’s standard model, including how energy is consumed and how they interact with the grid. As the data center industry prioritizes sustainability, there will inevitably be trade-offs, including in the design, the energy system, and the commercial structures.

Data centers are at the heart of the biggest issues of our time: sustainability, AI, the energy transition, and more. We look forward to continuing these robust and wide-ranging discussions with all the relevant parties across the data center space — including public sector players, like utilities, and hyperscalers — as we continue to explore how technology and creative approaches can address the big challenges in the field.

From left: Dian Grueneich, Stanford University; Wendy Stone, Silicon Valley Power; Ross Koningstein, Google Climate and Energy Research; Buddy Rizer, Loudoun County.