# AI's Demand for High-Density Racks Redefines Data Center Infrastructure

> As AI drives unprecedented changes in data center power and cooling needs, operators face the challenge of adapting infrastructure to support future workloads. Discover how innovative solutions are reshaping the industry.

**Source**: datacenterdynamics.com | **Published**: 2026-09-05 | **Type**: article

## Key Facts

- AI's demand for 1MW racks pressures infrastructure, risking obsolescence for older designs.
- NVIDIA's roadmap shows kilowatt increases; efficient token processing is crucial for profitability.
- Transitioning to 800V DC is essential; existing AC systems may hinder scalability and speed.
- Liquid cooling for power systems is vital; traditional cooling methods can't handle new heat loads.
- Collaboration among vendors is critical; no single supplier can meet diverse infrastructure needs.

## Summary

The data center industry is undergoing a significant transformation as artificial intelligence (AI) drives unprecedented changes in power and cooling requirements. With advanced compute platforms pushing rack densities to potentially 1 megawatt (MW) per rack, operators must rethink their infrastructure strategies to meet the demands of future workloads. This shift is critical as it impacts not only operational efficiency but also the long-term viability of data centers.

As AI applications evolve, they require more power and generate more heat than traditional systems. Current infrastructure, which was designed for older hardware, is being stretched to its limits. The decisions made today regarding power distribution and cooling technologies will shape the capacity of data centers to handle tomorrow's AI workloads. The industry is witnessing a surge in interest for higher-voltage direct current (DC) architectures, power sidecars, and specialized cooling technologies that can manage the increased thermal load associated with high-density AI systems.

In a recent discussion with Legrand’s Calvin Nicholson and Nick Schweissguth, it was highlighted that AI's influence has disrupted the traditional caution data center operators exhibit towards new infrastructure. For instance, NVIDIA's roadmap indicates a dramatic increase in kilowatt requirements, emphasizing the need for efficient processing to maximize profitability. However, this increased compute density can only be realized if the supporting electrical and thermal infrastructure is capable of keeping pace.

The Open Compute Project (OCP) typically operates at 48V to 50V DC, but delivering hundreds of kilowatts at these voltages necessitates extremely high currents, which can be problematic. Nicholson pointed out that the existing architectures do not scale effectively to meet these demands. As a solution, higher-voltage options, including 800V and potentially 1,500V DC, are being explored. This evolution will hinge on advancements in breakers, fuses, cables, and connectors, which must evolve in tandem with the increasing power demands.

Cooling strategies are also undergoing a paradigm shift. Historically focused on processors, cooling now must account for power-conversion equipment, which generates significant heat. Innovations such as rear-door heat exchangers for power sidecars and liquid cooling for busbars are becoming essential. Schweissguth anticipates a future where fully liquid-cooled IT stacks become standard, although the diverse requirements of various hardware platforms complicate the development of a universal cooling solution.

The uncertainty surrounding future compute platforms adds another layer of complexity. Operators may have a clear vision of their current systems, but the rapid pace of AI advancement means that future requirements could diverge significantly. One potential strategy is to overprovision power and cooling infrastructure to accommodate future upgrades. However, accurately predicting the necessary capacity remains a challenge.

As the industry adapts to these changes, the reliance on legacy systems will persist. Previous-generation GPUs and infrastructure will continue to hold value as overall demand for compute grows. The transition to higher-voltage DC will likely occur gradually, with many facilities integrating both AC and DC systems to manage the shift effectively.

The interdependence of various systems means that no single supplier can meet all the needs of a modern data center. Collaboration among design-build firms, integrators, IT manufacturers, and infrastructure specialists will be crucial in navigating this complex landscape. Vendors must also adapt quickly to real-time specifications, moving away from traditional product development cycles.

The evolution of data center infrastructure driven by AI signifies a critical juncture for the industry. As power demands escalate and cooling solutions become more sophisticated, operators must integrate these considerations into their planning processes. The future of data centers will hinge on the ability to harmonize power and cooling strategies, ensuring that facilities can support the next generation of AI workloads effectively. This integration will not only enhance operational efficiency but also position companies to capitalize on the growing demand for AI-driven services.

## Entities

- **Companies**: Legrand, NVIDIA
- **Technologies**: liquid cooling, two-phase direct-to-chip cooling, 800V DC, power sidecars, liquid-cooled busbars
- **People**: Calvin Nicholson, Nick Schweissguth
- **Organizations**: Open Compute Project

## Key Concepts

AI transformation in data centers, high-density computing, power distribution, cooling technologies, future-proofing infrastructure, voltage requirements, collaborative vendor ecosystems, thermal management

## Definitions

- **high-density computing**: A computing environment where the power and thermal demands of hardware are significantly increased, often requiring advanced infrastructure solutions.
- **liquid cooling**: A cooling method that uses liquid to absorb and dissipate heat from IT equipment, often more efficient than traditional air cooling.
- **two-phase direct-to-chip cooling**: A cooling technology that utilizes the phase change of a coolant to efficiently remove heat from chips, allowing for lower flow rates.
- **800V DC**: A high-voltage direct current standard that allows for efficient power distribution in data centers, particularly for high-density applications.
- **power sidecars**: Supplementary devices that convert and manage power distribution in data centers, often used to support high-voltage architectures.

## Use Cases

- Supporting AI workloads in data centers
- Implementing liquid cooling for high-density racks
- Utilizing power sidecars for voltage conversion
- Designing facilities for future AI infrastructure
- Adapting existing environments for higher voltage
- Collaborating with multiple vendors for infrastructure solutions

## Frequently Asked Questions

**What are the challenges of high-density AI computing?**

High-density AI computing presents challenges in power distribution and thermal management, as traditional infrastructure may not support the increased demands. Operators must rethink their cooling strategies and power architectures to accommodate these changes.

**How does liquid cooling differ from traditional cooling methods?**

Liquid cooling is more efficient than traditional air cooling as it can absorb heat more effectively, especially in high-density environments. It allows for better thermal management and can support the increased heat generated by advanced AI hardware.

**What is the significance of 800V DC in data centers?**

800V DC is significant because it allows for higher power distribution with reduced losses, making it suitable for high-density AI workloads. This voltage standard helps in scaling infrastructure to meet future demands.

**Why is collaboration among vendors important in data center design?**

Collaboration among vendors is crucial due to the complexity of modern data center requirements. No single supplier can provide all necessary components, so working together ensures that all aspects of the infrastructure are aligned and meet performance standards.

**What future trends should data center operators prepare for?**

Data center operators should prepare for trends such as increasing compute densities, the adoption of higher-voltage architectures, and the need for adaptable cooling solutions. Staying ahead of these trends will be essential for accommodating future AI workloads.

## Links

- [Read on Welcome.AI](https://welcome.ai/content/ais-demand-for-high-density-racks-redefines-data-center-infrastructure)
- [Original source](https://www.datacenterdynamics.com/en/marketwatch/planning-power-and-cooling-for-high-density-ai/)
- [NVIDIA](https://welcome.ai/company/nvidia): Featured company

---

Source: Welcome.AI | https://welcome.ai/content/ais-demand-for-high-density-racks-redefines-data-center-infrastructure