Data Center World 2026 conference highlights AI as pushing data center infrastructure against operational and capacity limits.

AI is forcing a fundamental reset in how data centers are designed, powered, and built. At Data Center World 2026, engineering leaders from Oracle Cloud Infrastructure, Nvidia, and Google described a critical shift: facilities are evolving from general-purpose IT environments into tightly integrated compute systems to support AI training and inference. This transformation is showing up across every layer of infrastructure—from power and cooling architectures to network design and construction timelines.

The shift begins at the workload level. Ram Nagappan, vice president of AI infrastructure at Oracle Cloud Infrastructure, explained that operators must now design for two fundamentally different AI patterns: large-scale training and distributed inference. Training workloads connect tens of thousands of GPUs in tightly coupled clusters, where latency and proximity matter. Inference workloads, by contrast, prioritize availability and responsiveness at a broader scale. These differences cascade through the facility. "You have to take both into account when you build the data center," Nagappan said, pointing to impacts on layout, resilience, and network design. The result is a more complex baseline: a single facility must support both tightly synchronized systems and distributed, user-facing workloads.

This complexity is compounded by rapid increases in rack density. Varun Sakalkar, distinguished engineer in Google's datacenter technology and systems group, said the industry has already moved past the thresholds that defined the last decade. Racks that once pushed 30–40 kW are now measured in hundreds of kilowatts, with designs approaching the megawatt range. The shift creates what Sakalkar described as a bimodal environment: traditional compute and storage infrastructure continues on a gradual density curve, while AI systems operate on a much steeper trajectory. As density rises, power availability—not compute—is emerging as the limiting factor. Sean James, distinguished engineer for energy systems at Nvidia, said operators are increasingly relying on on-site generation to accelerate deployment, but cautioned that these approaches are temporary. "Behind-the-meter power is a good stopgap," James said. "It's not the preferred long-term solution." Operators are working to secure grid-connected capacity while adding energy storage to manage increasingly volatile AI workloads, with sharp, dynamic load patterns that ripple beyond the data center itself. "You can see that impact all the way back at the power plant," James said. Energy storage is becoming essential to smooth those fluctuations, maintain power quality, and meet emerging grid requirements.

Cooling is undergoing a similar transition. Liquid cooling, once considered optional or niche, is now a baseline requirement for high-density AI systems. "Liquid cooling is here," Sakalkar said. "At this point, the conversation is about standardization." Operators must now manage hybrid environments where liquid-cooled AI systems coexist with air-cooled infrastructure. The industry is also confronting scaling challenges inside liquid systems themselves, from component supply chains to the number of connections required inside high-density racks. Water use is emerging as both a sustainability and operational risk, as James noted: "Data centers need to engineer out water where they can," pointing to the challenges of scaling evaporative cooling at hyperscale.

While systems grow more complex, deployment timelines continue to compress. Operators are responding by shifting work off-site and standardizing designs, relying on front-loaded design to ensure flexibility across GPU generations, increased use of prefabrication and factory integration to reduce on-site work, and modular architectures that can be assembled quickly. At the largest scale, the unit of design is changing again. Instead of optimizing individual buildings, hyperscalers are treating entire campuses as integrated systems. Sakalkar described this as a shift toward viewing the campus as a product, one that must balance flexibility, scale, and rapid deployment. Unlike traditional phased buildouts, many AI campuses are now deployed in large increments, with infrastructure and compute coming online in tighter synchronization. Across the panel, one theme remained constant: the traditional data center model is under strain, as AI is not just increasing demand but changing the shape of the infrastructure itself.