With enterprise investment in artificial intelligence exploding, sustainability conversations are increasingly focused on the growing intersection of AI, energy consumption and infrastructure resilience, especially as extreme weather continues to test the limits of global power systems.
The Southwest smashed heat records this March. The World Meteorological Organization just issued a report saying the last 11 years have been the hottest ever experienced by humanity. And now multiple scientists are warning that what’s coming next could be even more intense.
Spiking temperatures, unusually cold winter weather and other factors can stress grids and lead to power outages.
In a world that relies heavily on data and computing, how can enterprises prevent weather events and external grid disruptions from taking down data centers?Most start by deploying backup power supplies to build resilience.
But if you can understand events and control AI workloads, you can go even further to avoid grid instability, make your AI data centers more resilient while also improving energy efficiency so costs and resources aren’t overextended, and enhance your operational predictability.
Here’s how.
Energy grids have peaks and troughs, and AI workloads – whether for training or inference – tend to be spiky. Companies that build their AI data centers with grid-aware IT infrastructure can leverage these modern capabilities to identify the grid’s capacity at any moment in time and make informed, real-time decisions about workload and power usage.
Yet few organizations have full visibility into power usage or infrastructure performance. Less than half of organizations track average power demand data for their servers, and fewer than one-third can calculate work-per-energy efficiency metrics, according to Uptime Intelligence.
The data center of the future elevates observability as a critical capability, capturing and analyzing real-time data in an integrated fashion across smart, energy-efficient storage and other IT technologies in the rack, smart grids, renewable energy sources and advanced cooling. Understanding patterns, detecting trends and then mapping the impacts of events becomes foundational to intelligent data center management and optimization.
Enterprises can then schedule AI workloads in concordance with grid conditions. Imagine a heat wave hits, causing homes and businesses in the community to run their air conditioners continuously, or extreme winter weather blasts an area, prompting increased use of heating systems in a particular geography, maybe peaking at a particular time of day. Smart IT infrastructure can understand that the grid is under stress, assess whether to reschedule workloads within a different time window or whether to shift the workloads to other locations where the grid has greater capacity to meet their needs.
In the process, businesses using such intelligent IT infrastructure can help avoid grid overload, contribute to their communities and better align compute demand with available energy supply.
Building truly grid-aware AI infrastructure requires enterprises to bring together energy management software, building management software, cooling management software and even IT management software. This means shifting from simple monitoring to “single pane of glass” management within the data center, including across hybrid and edge sites. In fact, the recognition of this need is seen in the projected market growth for Data Center Infrastructure Management: an impressive 17.6% CAGR between 2025 and 2034, expected to approach a market size of $16 billion in that timeframe. But combining such complex systems is neither typical, historically, nor is it easy to do so seamlessly. It requires significant effort and expertise, across businesses and teams that may have always operated independently.
By partnering with experts to integrate these layers and interpret the complex data patterns that are observed, energy management can be interconnected with data from the rack to flag that data volumes are spiking, reveal additional energy or heat impacts of that event and provide guidance on how to mitigate the problem. Perhaps you need more cooling or dissipation, or need to plan for additional energy backup, for example.
Integrating different software layers into a cohesive whole can also enable enterprises to go beyond the walls of the enterprise and source information from the outside, such as details about emerging weather events. Now you can use massive amounts of information to predict such events’ impacts on your data centers. As noted, if one site is under stress, you may also opt to shift your workload or inference to another site, maintaining availability while avoiding unnecessary strain on localized power infrastructure.
The advantage of distributed data center design is that we can largely separate the locality of data center operations from business operations. Just because your business is in San Jose, California, for example, doesn’t mean your AI data center needs to be co-located in that city. Data can be moved to where power is available. To identify the ideal locations, start by looking for where power, land and other resources are available, and where other constraints are lower – including regulatory challenges, permitting process complexity and even community buy-in for the project. Take great care to follow all local ordinances and regulations – data sovereignty, for example, is becoming an increasingly important consideration, placing restrictions on where data is stored and how it is processed and moved. Overall, ensure that your IT infrastructure decisions deliberately and effectively balance performance, cost and sustainability goals.
Deploying alternative sources of power is another way to ensure you have the power you need. That’s why hyperscalers and other big AI data center investors are exploring and adopting small modular reactors, on-site generation based on sources such as natural gas, and solar, wind and battery power. This helps ensure businesses have access to the power they need at a time when communities are concerned about AI data centers’ impacts on grids and consumer energy bills.
Everyone is aware of AI’s incredible power to process large amounts of data to deliver rapid results and world-changing innovations. But lately the focus has been on AI’s power demands.
It’s no wonder why. Traditional data centers demanded about five to 15 kilowatts per rack, whereas AI data centers consume far more – 60 to 150+ kilowatts per rack.
But if you can design IT infrastructure that can tap into current energy information and adjust as needed, you can make your business and the grid more predictable, reliable and efficient, transforming resilience from a reactive safeguard into an intelligent, sustainability-driven advantage.
Simon Ninan is Senior Vice President of Business Strategy, responsible for developing and driving aligned execution of Hitachi Vantara’s business strategy for the short and long term, with the goal of maximizing customer and stakeholder value, while driving growth, innovation and market leadership.
