Fueled by innovations like smartphones, social media, ecommerce, and the cloud, data has exploded on a massive scale over the past two decades. To support this digital revolution, companies have constructed thousands of data centers around the globe. These facilities house endless rows of servers that store and process the trillions of digital bits that make up our emails, images, videos, and everything in between.
But breakneck data growth has pushed data centers into an untenable position. Their immense and ever-increasing appetite for energy clashes with escalating demands for companies to curb emissions, leading to increasing proliferation of intermittent renewables, though those aren’t a complete solution. Compounding the problem, amidst a broader trend of reshoring manufacturing to the United States, new advances have caused the explosion of machine learning and artificial intelligence, which require mind-boggling amounts of additional computing power. We’ve reached an inflection point where data centers’ unchecked expansion threatens communities, already-constrained electric transmission grids, and the planet. Hyperscalers recognize the need to do something different, but the scale of the problem is enormous.
The Core Sustainability Trap
The core issue stems from how large-scale data centers are engineered. Historically, they have been designed to guarantee 99.999% uptime to ensure continuous operation, regardless of the underlying customer needs. This level of reliability requires massive redundancy in power systems, including polluting diesel generators for backup power.
And the massive scale of these facilities translates into gigawatts of power consumption, with data centers generally tethered to the carbon-intensive grids that power them, and virtual renewable energy pledges insufficient to make up for their rapidly growing footprint.
The AI Compute Tsunami
If all that weren’t enough, AI workloads are now poised to drive energy demand to stratospheric new heights. Training a single AI model can produce over 626,000 pounds of carbon dioxide equivalent. That’s nearly five times the lifetime emissions of the average American car, including the manufacture of the car itself.(1)
AI-based services like chatbots require exponentially more compute cycles than standard web searches. As AI proliferates, its compute demands could exceed humanity's ability to build data center capacity — let alone doing it sustainably. The total energy demand of training AI models is projected to top tens of millions of terawatt-hours by 2027 — more than the use of many small countries.(2) Generating this colossal amount of power would emit millions more metric tons of CO2 annually than we already do.
Signs of Grid Stress Are Accelerating
Existing power grids cannot sustainably support data centers' endless growth. As demand swells in key hubs, signs of fragility have emerged.
Take Ashburn, Virginia, a major hub for data centers due to its proximity to the U.S.’s internet backbones. Dominion Energy, the local utility, announced in 2022 that it could no longer meet demand growth from data centers there. The massive capacity required for data centers is trapping energy that could serve other needs in the community.
And Dominion is not alone. In Ireland, the South Dublin County Council instituted a ban on new data centers due to overwhelming demand. Similarly, in Singapore, data center demand vastly outstripped supply, driving the government to implement a four-year moratorium on new data centers. Communities are recognizing that these facilities’ rapacious energy appetite starves other critical services like housing and hospitals of power.
Data centers reliant on diesel backup generators are unlikely to be able to secure enough fuel in the event of an extended emergency. For all their redundancy, critical failures lurk below the surface. And grids strained by supply-demand imbalances are increasingly implementing curtailment policies that force data centers offline during periods of peak demand.
These challenges illuminate the ultimate issues with the industry as it exists today. Diesel stockpiles cannot scale indefinitely. Backup generators can fail when needed most. Electric grids cannot spare endless capacity for digital infrastructure. Data center availability must be achieved through more innovative and efficient means.
Fundamentally Rethinking Data Center Engineering
All signs are pointing toward a reckoning for data centers. Tough questions loom about balancing sustainability and society's data needs. How much efficiency is required? Should growth be constrained? Who should sacrifice: communities, companies, or customers?
Data centers' unchecked expansion is colliding with hard realities. Unless we fundamentally transform how society powers and provisions computation, data-driven services risk becoming casualties of their own success — imperiling our grids and communities at the price of uninterrupted connectivity.
Addressing this systemic crisis demands reimagining data center engineering from the ground up. The challenges are complex, but the opportunity is massive. Data underpins innovation across every industry and society. We cannot put the data genie back in the bottle, nor should we. However, sustainable innovation must match exponential data growth.
With committed collaboration, a solution exists where human and environmental needs are not sacrificed at technology's altar. But we have little time to lose in charting a new course.
--