- AI workloads are pushing data centers beyond siloed management of facilities and IT resources
- Operators increasingly need orchestration layers that can coordinate infrastructure in real time, not just monitor it
- The payoff could be better capacity utilization, fewer performance bottlenecks and tighter control over cost and reliability
Data centers have historically operated as a collection of little kingdoms, with facilities teams managing power and cooling while IT teams wrangled servers and networks. But the rise of AI and high-performance computing is making this model harder to sustain. Rather than optimizing for their own distinct metrics, data center systems increasingly need to be orchestrated together.
What we’re talking about here isn’t too different from the idea of co-design between hardware and software that you might be familiar with. As GPU clusters push rack densities higher and workloads fluctuate faster, factors like compute placement, power availability, cooling capacity and even network behavior have become critical to squeezing the most out of constrained compute and power resources.
That shift is creating demand for a new kind of control layer, PADO AI, Hammerhead AI and Texas Instruments told Fierce.
The risk if data center operators can’t rise to the task? “Performance breaks first, followed by cost control,” PADO AI’s CEO Wannie Park said.
“Poor coordination rarely causes immediate catastrophic physical failure. Instead, it triggers a domino effect across the stack,” with cascading effects hitting performance, cost controls, reliability, cooling efficiency and then network capacity, he said. “We are moving rapidly toward a software-defined control plane where compute, networking, power and cooling operate as a single, holistic optimization domain.”
What does a data center orchestration layer look like?
At its core, orchestration means moving beyond visibility into active coordination. Data center operators already have plenty of telemetry coming from servers, GPUs, switches, power systems and cooling equipment. The harder problem is turning those siloed feeds into a control layer that can decide, in real time, how to place workloads, adjust cooling, avoid simultaneous power peaks and make fuller use of the capacity already inside the facility.
As Hammerhead AI CEO Rahul Kar explained, the goal is not simply to optimize cooling, compute or power independently, but to co-optimize them.
Today, there isn’t “one large system or large system of systems that is trying to bring everything together for a common outcome,” Kar said. “If you can coordinate, orchestrate across all these different silos and bring data and controls together, you can squeeze out not only more power, but also more compute from both newer infrastructure as well as the existing infrastructure.”
For instance, he added, you could use machine learning and AI to forecast when peak power utilization might happen within a data center and then use an orchestration system to control cooling, compute and anything else behind the meter as a means of managing through that peak.
But coordination depends on physical infrastructure that can measure and control what is happening inside the facility. Karthik Vasanth, VP and GM for Texas Instruments’ Data Converter Business Unit, said the company is looking at data center power and cooling “in a holistic manner,” from power efficiency and telemetry to temperature sensors, leak detection and pump controls.
"It has to be coordinated, because it is now a complete system solution," he said. "You cannot just say, 'I have the power, but…I don't know how to do the cooling or monitor the cooling.'"
To make a coordinated system work, Park said time-synchronized telemetry is needed across four key layers of data center infrastructure: compute (servers and chips), connectivity (switches and network), thermal management (cooling systems) and electrical (power distribution systems).
What roles do humans and AI have in data center orchestration?
Of course, all of this data would be too much for a human to sift through and react to in real-time. So, AI and automation have a role to play here.
“Automation and AI should act as the high-speed nervous system of modern data centers—handling sub-second telemetry, predictive adjustments, and real-time co-optimization across compute, power and cooling,” Park said.
But that doesn’t mean humans disappear from the loop. Park said people still play a crucial role in setting strategy, governing risk, establishing policies across different infrastructure domains and stepping in when unusual failure scenarios arise. Humans are also needed to define the guardrails and weigh trade-offs that AI systems may not be able to fully measure or resolve on their own.