As AI reshapes global computing needs, future-proofing data centers requires smarter energy use, sustainable cooling, and realistic long-term planning for resource demands.
A new study from McKinsey Global finds that organizations are now implementing AI into the bottom line. From redesigning workflows to connecting AI governance to senior leadership, AI can be transformative, but it also comes with new challenges. Companies need to take steps to address the risks of AI while also leveraging it to its fullest. One major area that requires attention is how to prepare data centers to handle AI-driven workloads.
Water, data centers, and AI workloads
Both generative AI and data center management already have one key factor in common – they both guzzle the world’s water supply. Data centers have already begun tapping into the world’s supply of fresh drinking water, which accounts for only 0.5% of Earth’s water (3% when factoring in non-potable fresh water). Data centers use water to cool processor chips, preventing overheating and damage. As data centers handle AI requests, they produce more heat and need more water.
A large data center can use up to 5 million gallons of water each day. This is equivalent to the water use of a town with 10,000 to 50,000 people. A medium-sized data center consumes up to 110 million gallons of water each year, comparable to the annual water use of approximately 1,000 American households. Nearly half of all US data centers are powered by water-intensive power plants that use “swamp cooling†techniques. Warm air is drawn through damp pads, leading to around 80% evaporation and water loss. The rest becomes salty discharge, sent to wastewater treatment plants. For context, residential water use accounts for just 10% of evaporation.
Around 20% of data centers in the United States rely on water sources that are already classified as “moderate to high stress†due to drought, local consumption, pollution, and other factors. Meanwhile, nearly half of all US servers are located within water-stressed regions.
AI is only going to exacerbate the problem of water management, both worldwide and in the United States. The global annual water consumption for generative AI is projected to range from 4.2 to 6.6 billion cubic meters by 2027. This is as if a country that needed 4 to 6 times the amount of water as Denmark does each year suddenly appeared on the scene, drinking up the world’s resources.
Even with all of these facts in place, not many companies have created or shared publicly plans regarding resource management. In a study of 122 companies that manage data centers, only 16% have filed plans that address water risk.
Managing the AI workload
Overuse of AI will lead to massive depletion of world resources, mainly fresh water, which is the most critical human resource of all. Companies must take concrete steps today to address the risk of data centers handling increased AI-driven requests, processing and storage. One option is segmentation. Creating a contained segment within a data center for AI will require an organizational overview, but it can save power in other areas while handling AI requests when needed. Companies should consider analyzing their energy consumption and data management needs, and invest accordingly. Other options involve:
Investing in new servers
Next-generation servers are better equipped to handle the massive flow of data from generative AI, and may save energy from smoother processing. While investing in new servers comes with an upfront cost, doing so can create a better, more efficient workflow for a while to come. You might consider your company’s growth projections, data storage needs, budget, and a realistic overview of how your company plans to utilize AI over the next 5 to 15 years.
Running lower resolution
While scientific research often requires higher-level calculations, AI workloads can run at half-precision (FP16) for most other applications. Even quarter-precision may be possible for businesses that primarily use AI to handle customer service responses or inquiries. Consider what your business needs before running full 64-bit calculations through generative AI programs.
Considering edge computing
Edge computing is low-latency, privacy-sensitive, and also highly efficient with larger workloads. Edge computing utilizes a network of sensors and IoT devices to keep sensitive information closer to users. It may be favored by companies with sensitive information, such as health care data or financial accounts, due to its proximity benefits and swifter processing times.
Working with partners who plan for reality
Choose data center partners who think long term, not just large scale. Ask for their ESG and water management plans, and review how they handle drought risk, wastewater, and local regulations. Transparency matters, and more than $64 billion in data center projects have already been delayed or blocked because of water concerns.
Local opposition comes from both sides of the political spectrum, with 55% of Republican and 45% of Democratic districts pushing back on data center growth. Partnering with facilities that ignore sustainability can cost your company time, money, and reputation. Before expanding AI workloads, make sure your infrastructure partners have realistic strategies for water use and resource management. A little due diligence today can prevent major losses tomorrow.
About NetworkTigers
NetworkTigers is the leader in the secondary market for Grade A, seller-refurbished networking equipment. Founded in January 1996 as Andover Consulting Group, which built and re-architected data centers for Fortune 500 firms, NetworkTigers provides consulting and network equipment to global governmental agencies, Fortune 2000, and healthcare companies. www.networktigers.com.
