Using the power of innovation to impact the world around us
SIGNALS+ NEWSLETTER SUBSCRIPTION
Stay updated and leverage Signals+ latest insights, information and ideas on Connectivity, Digital Health, Electrification, and Smart Industry.
Thank you for subscribing to ADI Signals+. A confirmation email has been sent to your inbox.
You'll soon receive timely updates on all the breakthrough technologies impacting human lives across the globe. Enjoy!
Close
Close Details
ADDRESSING THE DATA CENTER COOLING CHALLENGE
KEY TAKEAWAYS
|
Data centers are undergoing rapid transformation and expansion driven by surging demand for data, AI, and complex computational workloads. They face two critical challenges: efficiently powering and cooling the ever-increasing hardware density within server racks.
Modern AI, especially those built for AI and machine learning, packs massive computational power into smaller footprints. More transistors, denser GPU clusters, and high-performance accelerators result in higher power consumption and intense heat in confined spaces.
The limiting factor for data centers has shifted from floor space and power to heat dissipation. Across the industry, from traditional operators to hyperscalers, this thermal inflection point is forcing a fundamental rethink of how systems are designed, managed, and optimized.
THE THERMAL WALL
Density is increasing rapidly.
Traditional racks operate at tens of kilowatts. AI workloads such as large language models like ChatGPT require 40kW to 60kW of GPU rack power. Emerging hyperscale “AI factories,” purpose-built for AI training and inference, are already exceeding 100kW, with future designs aiming for 500kW to 1MW per rack.
Air cooling doesn’t scale.
Air is an inefficient heat-transfer medium. At higher computational densities, it fails to remove heat fast enough, forcing unsustainable increases in airflow, fan power, and rack-level complexity. Conventional CRAC/CRAH systems become impractical and cost-prohibitive above ~50kW to 100kW per rack.
Thermal risks compound.
Insufficient cooling leads to throttling, failures, shorter component lifespan, and higher operational costs. Heat, not space or power, is becoming the primary limiter of performance, uptime, and capacity.
Every watt spent fighting heat is a watt NOT used for computing.
RETHINKING POWER AND COOLING
As processor power increases and chip density rises, heat output exceeds what traditional air-cooling systems can efficiently or cost-effectively manage. This limitation is driving the adoption of liquid cooling, a fundamental redesign of thermal management that dissipates heat orders of magnitude more effectively than air. Cooling now evolves alongside power delivery, rack layout, and airflow, an integrated approach that allows hyperscale AI data centers to surpass previous density limits.
The liquid solution.
Liquid’s superior thermal conductivity extracts heat directly at the source. While introducing added complexity and cost, the transition is underway. Hyperscalers are building 100% liquid cooled facilities. Existing sites are adopting hybrid approaches, liquid for high-density racks with air cooling elsewhere. Liquid cooling is essential not only for today’s heat loads but also for sustaining AI-driven workload growth across all data center scales.
Preventing heat generation.
Operators must balance removing heat with preventing heat generation, a consideration extending beyond thermal management to the entire power delivery chain. The bigger lever is minimizing heat at the source, driving higher power conversion efficiency, reducing resistive losses, and optimizing voltage regulation, so less energy is wasted as heat in the first place, easing cooling demand before it even becomes a problem.
Powering the future: 800V DC distribution.
AI scale racks challenge traditional power systems: 12V, 48V, and 415VAC rely on massive copper runs, bulky power supply units, and inefficient conversion chains that increase heat. The industry is shifting toward 800V DC distribution, which reduces conductors, decreases resistive losses, and simplifies delivery.
Analog Devices supports 800V DC architectures with hot swap controllers, high efficiency DC-to-DC conversion, power monitoring, and advanced protection for stable, efficient delivery.
INTELLIGENCE AT THE RACK
Advanced monitoring tracks voltages, currents, and temperatures in real time, enabling precise resource allocation. Running cooling pumps at full capacity when it’s unnecessary wastes energy. Accurate metering and precision components, like low-noise amplifiers, measure current-sensing shunt signals, which are vital to efficient power delivery and thermal management.
Operators run systems at approximately 99.95% power level rather than 100% capacity to maintain reliability. ADI monitoring solutions track these power levels in real time, enabling operators to dynamically adjust load structure to balance performance with system longevity.
As AI advances, fueling the demand for both more power and more powerful data centers, it is simultaneously providing the intelligence to manage those demands efficiently. AI amplifies precision by acting as a smart assistant, processing sensor data to identify patterns, predict failures, and automate adjustments. This enables operators to detect anomalies before downtime, and dynamically adjust cooling based on real-time workloads, transforming reactive maintenance into proactive management.
At the rack level, specialized components make advances possible:
- Hot swap controllers and protection ICs manage live power insertion/removal, limiting inrush current, preventing spikes, detecting faults, and ensuring safe operations—critical for 800V feeds and megawatt loads.
- Voltage regulation and DC-to-DC conversion down-convert 800V DC to voltages required by GPUs, CPUs, memory, and components using high-efficiency regulators and multiphase controllers. Analog Devices solutions optimize conversion for minimal losses and thermal load.
- Digital telemetry and thermal monitoring provide real-time visibility and control in high-density environments, enabling operators to detect issues and manage thermal boundaries.
- Battery backup and capacitive backup systems monitor voltage on backplanes feeding IT racks. When power fluctuates or disappears, these systems provide power or absorb charge as needed. ADI’s battery management solutions, originally developed for automotive applications, ensure uninterrupted operation during transitions.
These components enable stable high-voltage distribution and tight power/thermal control, making high-density, liquid-cooled racks practical, safe, and manageable.
THE BUSINESS CASE FOR INTELLIGENT DATA CENTER INFRASTRUCTURE
|
New architecture and advanced technologies enable measurable improvements for data centers.
- Precision sensing and AI-driven maintenance help reduce downtime and extend equipment lifespan.
- Intelligent control systems promote energy efficiency by aligning resources with demand.
- Liquid cooling allows for unprecedented computational density within existing space.
Organizations adopting these approaches typically experience lower operational costs, fewer failures, and progress toward sustainability goals with benefits that scale across operations.
PAVING THE WAY FOR TOMORROW
The future data center will be a finely tuned ecosystem where advanced components, from power management and sensing to optical connectivity and battery management, work together seamlessly. This integrated approach addresses today’s challenges while supporting tomorrow’s computational demands. Whether retrofitting legacy facilities or designing greenfield deployments, liquid cooling enables data centers of all scales to remain vital hubs of digital innovation.
Data center transformation is more than just about thermal management and power efficiency; it’s about unlocking the next frontier of possibility.
References
1 Zachary Skidmore. “Microsoft Study Finds Liquid Cooling Can Cut Data Center Emissions by up to 21%”. Data Centre Dynamics, May 2025.
