Data center monitoring and measurement technology is evolving at a rapid pace. This is coming in response to three primary forces:

• AI and GPU spending is at an all-time high. The latest generation of data center infrastructure requires far more sophisticated monitoring and management capabilities.
• Compliance challenges are mounting related to power usage effectiveness (PUE) optimization, and there is a greater need for measurement and accountability of data center spending. If the amount being spent is going to increase significantly due to AI and GPUs, operators and stakeholders need to be confident they can measure the return on these investments.
• Unprecedented power densities and load volatility are pushing the data center thermal chain to the limit. Traditional mechanical systems, controls, and monitoring protocols were designed for stable, predictable IT loads, not the rapid spikes and troughs expected from “AI factories.”

Expect, therefore, to see more data centers designed and operated using digital twins. These twins are virtual representations of an entire data center or the systems running within it. Designers are using them to digitally configure the data center to see how best to fit everything together before anything is built. Once operating, the digital twin shows how each component and server is functioning, highlights problems, and helps minimize downtime. Further, the data center manager can simulate changes using the digital twin to see how best to add new equipment, introduce liquid cooling, or add more GPUs.

These twins will also apply to cooling systems. Sudden spikes in heat load will generate alerts. In advanced systems, automated controls will trigger AHUs to provide more cooled air, while liquid cooling systems will be directed to pump more fluid to rear-door heat exchangers and direct-to-chip (DtC) cold plates.

Expect, too, broader deployment of AI-based dynamic optimization of cooling monitoring and control systems as pressure to lower PUE intensifies. Digital twins and AI systems will be fed immense amounts of data from a dense network of sensors, thermal cameras, and smart valves operating throughout the cooling system, power infrastructure, and computing hardware. These sensors will continuously transmit information from the rack, the coolant loop, the PDU, and individual components in real time.

For example, sensors can be embedded in cooling loops to monitor temperature, pressure, coolant quality, and flow rates. Their data can be used to greatly enhance predictive maintenance capabilities. AI may eventually evolve to the point where it can spot the early stages of wear or failure and adjust settings to resolve the problem. Central AI systems at the heart of operations will offer predictive models that forecast load and heat events and can dynamically adjust operations based on heat load, temperature, and system performance.

Bridging the Gap to the AI Data Center

A few hyperscalers and large colos may be able to implement such technology directly into the new AI factories that are under development. However, most existing data centers are still far from this vision.

What steps should they be taking to move toward the AI data center of tomorrow?

A good place to start is by adding sensors throughout existing cooling and power infrastructure. Immersive monitoring and optimization software can tie into these sensors to monitor, manage, and maximize the performance of power and cooling infrastructure across data center environments. With such a network in place, operators can gain real-time visibility into thermal and power conditions through intuitive 3D visualizations of cooling systems, power infrastructure, and the entire data center space.

This type of visibility can help data center teams identify thermal risks, optimize cooling performance, improve capacity planning, and make more informed decisions as infrastructure and workloads evolve. Just as importantly, it establishes the foundation of accurate, real-time data that AI-based optimization depends on to provide meaningful conclusions and recommendations.