Loading page…
The technology is beginning to optimize cooling, power allocation and incident response at infrastructure scale. The limiting question is not whether AI can find a better action, but which actions operators will permit it to take alone.
Listen to this story
Digital Realty’s data-center AI rollout is moving from pilot to scaled operations, with more than 10 facilities using OPDaaS and about 30 additional sites targeted by the end of 2026. The practical use case is optimization—especially cooling—not autonomous control: Phaidra is tuning liquid cooling at IAD51, while operators retain accountability for risky changes.
Phaidra’s platform adjusts IAD51’s secondary liquid-cooling loop to computing demand rather than running at full capacity.
Digital Realty’s OPDaaS centralizes sub-second power and cooling telemetry and exposes it through APIs to internal and customer tools.
AWS is using generative-AI agents for network incident investigation and some routine fixes, a narrower autonomy step than facility control.
Digital Realty has deployed its operational AI platform in more than 10 data centers and plans to add roughly 30 more by year-end. The rollout shows where data-center AI is becoming concrete: systems ingest fast-moving facility data, flag waste or failures, and give operators a basis for action rather than taking over the building.
Digital Realty’s Operational Data as a Service, or OPDaaS, aggregates sub-second telemetry from power, cooling and other building systems in a central store. APIs make the data available to the company’s tools and to customer tools, creating a control layer that can identify conditions such as air-system filters forcing fans to work harder than necessary.
At its IAD51 facility in Northern Virginia, Digital Realty is deploying Phaidra’s AI platform to tune a secondary liquid-cooling loop to actual computing demand instead of operating at full capacity by default. The company said it saved 17,800 MWh in 2025 and attributed part of its efficiency gains to AI-driven optimization; that is a company-reported outcome rather than an independent measure of the software’s contribution.
Digital Realty says OPDaaS is deployed in more than 10 facilities.
The company plans to bring about 30 additional data centers online by the end of 2026.
Digital Realty said it saved 17,800 MWh in 2025 and credited AI-driven optimization in part.
AWS is applying generative-AI agents to network operations, where they correlate telemetry from multiple monitoring systems, investigate incidents and identify root causes. For some routine problems spanning services, compute infrastructure and networking, AWS says the systems review incoming tickets against historical patterns and resolve issues without human intervention.
That is a narrower form of autonomy than handing a facility to an agent. AWS also uses generative-AI software to place servers in racks more efficiently and reduce stranded power, meaning available electricity that cannot be used because of how capacity is arranged.
Schneider Electric’s EcoCare service-dispatch system recommends a technician, likely replacement parts and an urgency level, but human dispatchers make the final choice. Digital Realty likewise has a network-operations-center team watching its systems, even as it expects to allow more algorithmic decisions in time-critical situations over time.
That restraint is still typical. Hyperscalers and the largest colocation providers are furthest ahead in deploying data-center AI, while full autonomy remains rare. The immediate shift is toward automating bounded tasks—diagnosis, simulation, dispatch recommendations and routine fixes—while people retain responsibility when the action carries operational risk.
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.