# How Data Centers Are Using AI to Run Cooler and Smarter > Source: > Published: 2026-08-27 09:00:00+00:00 # How Data Centers Are Using AI to Run Cooler and Smarter Analysts weigh in on where the industry is headed, as Digital Realty, AWS, DPR Construction, and Schneider Electric demonstrate how they’re leveraging AI. Data center operators are racing to build capacity for AI workloads, and as they do, they are increasingly deploying AI to run their own facilities more efficiently. From cooling and power optimization to predictive maintenance and incident response, AI is moving from pilot to production across the industry. While full autonomy remains rare today, AI is delivering measurable gains in energy use, uptime, and operational speed. Hyperscalers and the largest colocation providers lead with in-house tools, midsized operators are piloting targeted use cases as operational pressures mount, and many smaller environments are still evaluating adoption. This article examines how Digital Realty, AWS, DPR Construction, and Schneider Electric are applying AI today and where adoption is headed next. ## AI in the Data Center: What’s Working Today Analysts say operators are still in the early stages. Today’s deployments focus on analyzing data to predict problems before they occur, optimizing [cooling and power use](/build-design/ai-transforms-data-centers-into-power-and-cooling-plants), improving IT resource utilization, and providing technicians with better information to fix issues. In some cases, AI resolves problems autonomously, but humans generally retain final decision-making. “AI is being used in discrete packages to do little bits of the work,” said Roy Illsley, Omdia’s chief analyst for IT operations. Operators want to eventually optimize entire data centers in real time, Illsley said, but that’s not happening yet. “We are going to move to a world where a lot of the specific tasks are done by AI agents, and then the humans will be connecting those tasks and overseeing decisions,” he said. Hyperscalers and the largest colocation providers are furthest ahead, often building their own tools. Some midsized operators are experimenting as they feel operational pains, while most enterprises and smaller data centers have yet to adopt AI, Illsley said. ## Digital Realty: OPDaaS, Cooling Optimization, and Measured Savings Digital Realty has applied machine learning in its facilities for more than five years and has used generative AI since it arrived about three to four years ago, said Chris Sharp, the company’s CTO. The company operates more than 300 data centers worldwide, serving customers from hyperscalers to enterprises. Its in-house Operational Data as a Service (OPDaaS) platform is a curated set of AI tools that pulls sub-second telemetry from power, cooling, and other building systems into a central data store. The platform exposes that data through APIs so both Digital Realty’s own tools and customer tools can analyze it and tweak power and cooling to improve efficiency and support predictive maintenance, Sharp said. For example, AI can detect when filters for air-cooled systems are getting clogged, forcing fans to run at or near 100% unnecessarily. Proactively cleaning or replacing filters can deliver double-digit percentage efficiency gains, he said. Digital Realty is also deploying Phaidra’s AI platform at its IAD51 data center in Northern Virginia, where Nvidia is a customer. The tool provides visibility into the secondary liquid cooling loop, allowing cooling to match actual compute demand rather than defaulting to full capacity, Sharp noted. The company has deployed OPDaaS in more than 10 data centers, with plans to bring about 30 more online by the end of 2026. According to the company’s 2025 Impact Report, Digital Realty grew its portfolio by 34% while increasing water usage by only 3% -- gains Sharp credited in part to OPDaaS and AI-driven optimizations. “That’s a huge delta that we couldn’t have achieved without generative AI being inside our systems, monitoring pumps, filters, and the full spectrum of infrastructure,” he said, adding that the company saved 17,800 MWh in 2025. Today, humans make operational decisions based on what AI systems discover, but Sharp expects that to change over time, with the company gradually allowing some decisions to be made more autonomously, especially for time-critical situations. Digital Realty currently has a team in its network operations center watching the system. “You’ll start to see a little bit of those decisions being made with the oversight of a human. That’ll be our first step in that foray,” Sharp said. “But I do see a future, because of the complexity of the infrastructure, you’re going to have to allow a lot of decisions to be made more algorithmically and in a more autonomous fashion.” ## AWS: AI Agents for Network Operations and Stranded Power Reduction AWS is using software powered by generative AI to optimize server placement in racks to reduce stranded power – available energy that goes unused. The company is also using AI and agentic workflows across its network operations to manage infrastructure at scale. AWS’s network is highly automated, but there are cases where automation can’t determine the correct action, so systems escalate to engineers, the company said. The goal is to use AI agents to investigate, collect, and correlate across different data sources. For incident response, AI agents automatically investigate network issues by correlating telemetry from multiple monitoring systems to identify root causes in seconds rather than minutes, significantly reducing the time it takes to diagnose issues for its customers, said Zak Islam, AWS’s director of network observability and automation. For routine operational issues across services, compute infrastructure, and networking, AI reviews incoming tickets, compares them with historical patterns, and, in many cases, fixes the problems without human intervention, freeing engineers to focus on more complex or novel problems, he said. AWS also uses AI to monitor its fiber-optic infrastructure, automatically reaching out to vendors to speed up repairs, Islam said. During network buildouts, the company uses AI to flag configuration errors and route them to the right team, and engineers increasingly use natural language to query network telemetry data rather than manually inspecting dashboards. Overall, AWS has leveraged machine learning and automated reasoning for more than a decade. “With the proliferation of LLMs, we are now able to use these systems at unprecedented scale across a wide range of technical and business problems,” Islam said. ## DPR Construction: AI-Driven Pre-Construction and Site Robotics DPR Construction, a national contractor in Santa Clara, California, that builds data centers among a range of facilities, has ramped up its use of AI to help staff make better decisions on construction projects, said Max Lares, a DPR project executive. For the past two to three years, the company has used AI for data analytics during early pre-construction, drawing on benchmarking data from past builds, such as site conditions, cost models and construction sequencing, to predict staffing needs, project duration, and sequencing for new builds, and to help tenants predict cost and value. The team runs simulations to plan construction sequencing from foundations and steel erection to mechanical, electrical, and fire protection systems, Lares said. That helps DPR determine whether to start from the middle of a building or from one end to the other, he noted. DPR is also piloting robots that walk job sites overnight to take photos, so customers can keep tabs on progress. With robots deployed, construction crews no longer have to move cameras and take photos themselves during work hours. Lares said the robots also produce better images. “Robots can just go and walk the site and take the photos for us at a time when there’s no one on the job site,” he said. “It has much better views of the entire area of construction because there’s no construction taking place.” Lares added that the goal is to become more efficient and effective in delivering services for its customers, but the company will not replace human judgment. “AI is a great tool for us, but we still remain responsible and accountable for our own decision making,” he said. ## Schneider Electric: Digital Twins, DCIM, and Governed AI at Scale Schneider Electric, which supplies power, cooling, and other data center infrastructure, uses AI in its products and internal operations. Data center operators can use the company’s ETAP electrical power system modeling software and AVEVA industrial operations software with Nvidia technology to create [digital twins](/data-center-software/nvidia-dassault-partner-for-digital-twin-ai-platform) of power and cooling infrastructure, the company said. AI and digital twins enable operators to run simulations and predict how systems will behave before changes are made in the real world. Schneider Electric’s EcoStruxure IT software also allows operators to continuously monitor infrastructure, using AI for remote diagnostics, proactive recommendations, and predictive maintenance. The company uses AI for service dispatch as part of its EcoCare services, recommending the right technician, likely replacement parts, and urgency level, while human dispatchers make final decisions, said Steven Carlini, the company’s chief advocate for data centers and AI. “There’s a human in the loop,” he said. “Right now, as with most AI, it’s not fully agentic or autonomous.” On the business side, Schneider Electric’s AI journey dates back eight years, but the effort stayed largely in pilots until about a year and a half ago, when the company began pushing to scale AI across its business, said Jennifer Swen, Schneider’s vice president of operational excellence and AI. For example, the company launched an agentic system that processes customer quote requests, cutting the process from 62 steps to 14 and shortening turnaround time by about 95%, Swen said. The company also uses AI to help its sales team prepare for customer meetings, pulling account history, open opportunities, and pipeline data into a single summary. All internal AI tools run on a centrally governed cloud platform, with strict oversight from the company’s chief AI officer. "We put our AI solutions through a very robust risk and governance process," Swen said. ## What’s Next: IT/OT Convergence and Agentic Control AI is more mature on the IT side of the data center, which includes servers, storage, and applications, than on the operations side, which covers power distribution, cooling, and other facilities infrastructure, Illsley said. On the IT side, AI is largely deployed through AIOps tools that ingest logs, traces, and telemetry to detect issues. “You’ve got all of this information being written and pumped out,” Illsley said. “The tools are picking the signal out of all that noise, and bringing it to the attention of the operators, so the operators can take it further.” That might mean flagging database corruption, a full disk, or a network problem before engineers discover the issue, he noted. The OT side is catching up. Vendors such as Schneider Electric and Siemens are building AI into their data center infrastructure management (DCIM) systems to monitor power distribution, cooling, pumps, valves, and chillers, and to diagnose anomalies, such as a blocked chiller filter, he said. In Illsley’s view, the next logical step is to bring IT and OT data together in a shared layer so AI can see both sides and help operators make smarter, more holistic decisions than either side can make on its own. The IT and OT tools would remain separate. That bridge between IT and OT is starting to take shape. Zoe Roth, senior research analyst at 451 Research, part of S&P Global, said a new class of “agentic analytic” startups sits atop existing cooling and power systems, orchestrating cooling and thermal optimization to reduce energy use. One startup, Phaidra, also streams telemetry from IT systems, giving operators visibility across building management, power, cooling, and IT. Its reinforcement learning agents then use that data to generate commands for the cooling infrastructure, helping to keep equipment at a more stable, efficient temperature, Roth explained. ## Will Data Centers Be Fully Autonomous? Even with AI’s advancing capabilities, analysts said fully autonomous data centers remain a remote possibility – not because the technology isn’t capable, but because of trust. “Would the AI be fully capable of running a data center? Almost certainly – yes,” Illsley said. “But would we trust it? Probably not.” Dan Thompson, research director at 451 Research, part of S&P Global, said the calculus changes for data centers in places people can’t reach easily, like the North Slope of Alaska. Those proposed facilities will have to operate autonomously or at least maintain strong connectivity to human teams for troubleshooting, he said. The same goes for [data centers in space](/build-design/breaking-points-2035-a-data-center-space-odyssey). “It’s not terribly practical to send an astronaut into space every time there are issues,” Thompson noted. For an ordinary data center, though, humans are likely to remain in the loop for the foreseeable future. “My perception is that, at least right now, the average data center operator just wants to be armed with better information about what to do and perhaps the best course of action to take,” Thompson said. Someone still has to be accountable for what happens onsite, even as AI agents take on more specific tasks. “You are responsible and accountable for that data center, not the AI agent that’s flying around at 5 million miles an hour,” Illsley said.