Inside the Rack: How JetCool and Flex Deliver Turnkey Racks for AI Cooling

Ally Lynch

7 months ago

Data center operators are facing unprecedented AI compute demands forcing them to rethink infrastructure, starting at the rack. No longer a passive enclosure, the modern rack is evolving into a complex, integrated system where power delivery, cooling systems, and component compatibility must be co-engineered from day one.

Barriers to Scaling AI Infrastructure

While the vision for AI-ready infrastructure is clear, getting there isn’t so simple. Despite growing demand and rapid innovation, scaling AI infrastructure faces systemic hurdles across power, cooling, integration, and global deployment.

Thermal Headroom Is Running Out: Traditional air cooling can’t keep up with AI workloads, leading to throttling and reduced hardware lifespan. Liquid cooling is essential but requires precise engineering and infrastructure.
Power Delivery Isn’t Keeping Up: AI racks pushing 600kW–1MW strain legacy AC systems. Without DC distribution and integrated energy solutions, scalability and efficiency suffer.
Global Consistency is Difficult: Regional differences in vendors, standards, and site capabilities turn each deployment into a custom project—raising costs and complexity.
Integration Complexity: The tech exists, but the ecosystem lacks standardization. Blurred lines between IT and facilities—especially with cooling—cause delays and risk.

The Result: Integration Paralysis Sets In

Designing and deploying AI infrastructure at scale is no small task. Enterprises face a rapidly evolving hardware landscape, with new platforms arriving every six months. That pace makes it essential to invest in systems that can adapt over time, rather than risk being locked into a design that quickly reaches its limits.

Future-ready racks call for compatible components, streamlined vendor coordination, and validated performance across diverse environments. Without common standards or tight supply chain integration, projects can take longer than expected and require additional design cycles.

Instead of slowing down progress, organizations are increasingly looking for integrated solutions that minimize complexity, reduce time to deployment, and keep teams focused on scaling infrastructure rather than troubleshooting integration.

The Evolution of Rack Design in the Age of AI

The limitations of legacy integration models have made one thing clear: the rack can no longer be an afterthought. Historically, racks served as passive enclosures—standardized frames to house compute. But in the age of AI, it must evolve into a high-performance platform. The thermal and electrical profile of the rack has changed dramatically. What once spanned an entire row is now concentrated into just a few square feet. This compression of compute into high-density footprints has forced a fundamental rethink of rack architecture—not just in terms of performance, but in how racks are designed, sourced, and deployed.

That rethink is now reshaping operational models. Previously, operators could assemble racks from a mix of vendors—servers from one, power distribution from another, cooling added as needed. But that piecemeal approach is not viable. The demands of AI—extreme power density, precise thermal management, and spatial constraints—require tight integration across all rack subsystems.

And we’re already seeing this shift. Platforms like NVIDIA’s MGX rack and Mount Diablo reflect a new design philosophy—where racks are no longer assembled as a collection of parts but pre-integrated ecosystems. As demands grow, the question isn’t if infrastructure should evolve, but how that evolution can be delivered at scale.

The Turnkey Rack Alternative

To overcome these bottlenecks, many operators are turning to turnkey rack platforms—fully integrated systems built and validated at the factory. These racks arrive with cooling, power, and telemetry pre-installed, engineered to function as a cohesive unit.

This model offers several advantages:

Faster time-to-value: Ready-to-operate systems reduce deployment timelines.
Reduced complexity: Upstream integration minimizes vendor coordination and on-site troubleshooting.
Predictable performance: Factory validation ensures systems function as intended, reducing on-site troubleshooting.
Versatile Device Integration: Designed to accommodate 3U, 4U, and 6U height units.
Scalability: Standardized designs support consistent replication across diverse global sites.

By addressing integration challenges at the source, turnkey racks offer a more streamlined and scalable path for deploying AI infrastructure.

Inside the JetCool + Flex Vertically Integrated Racks

Flex and JetCool, now a Flex company, meet the market where it needs it most: at the intersection of thermal, power, and operational complexity. Their vertically integrated rack platform is designed to address the demands of AI infrastructure by combining liquid cooling, power delivery, telemetry, and installation into a single, factory-engineered system.

Smart Cooling at the Chip Level

JetCool’s SmartPlate™ cold plate technology applies liquid cooling directly to high-power devices, using jet impingement to achieve low thermal resistance and support workloads exceeding 4kW TDP in a single socket. This precision cooling helps maintain performance, extend hardware lifespan, and maximize utilization.

Warm Water Cooling, High Delta-T Efficiency

Unlike traditional liquid cooling solutions, JetCool’s architecture supports warm-water cooling (>60 °C inlet) and high delta-T operation. This enables efficient heat reuse, eliminates the need for energy-intensive chillers or evaporative towers year-round, and simplifies facility integration.

At the rack level, the SmartSense CDU orchestrates fluid delivery with built-in telemetry, monitoring, and redundancy. It scales to 300 kW per rack and 1.8 MW per row, supporting hyperscale footprints without custom plumbing or exotic infrastructure.

Power at Scale: Integrated 800Vdc Architecture

Flex delivers advanced power solutions to accelerate the deployment of 800Vdc architectures, meeting the rising demands of AI infrastructure and AI factories.

Its high-density power shelf system provides up to 33 kW per shelf within a compact, industry-standard footprint. Designed with six power supply units in a 3+3 redundant configuration and an intelligent power shelf controller, the system achieves peak efficiency of 97.5% at half-load, ensuring reliable and energy-efficient DC power distribution.

Telemetry: Visibility from Day One

Factory-integrated sensors provide real-time visibility into thermal, power, and fluid metrics—enabling predictive maintenance, fault response, and SLA management without external systems.

Designed for Scalable Deployment

Pre-wired and thermally validated, the rack is built to meet sustainability benchmarks and can be replicated globally through Flex’s manufacturing network.

These turnkey racks are not proof-of-concept pilots; they’re production-grade platforms already being used to power AI clusters for hyperscalers and advanced enterprise environments.

Closing the Gap Between AI Vision and Infrastructure Reality

As AI permeates across industries, its infrastructure must evolve in lockstep. The era of field-integrated racks, mismatched components, and delayed deployments is over. Now, operators need speed, standardization, and sustainability from day one.

JetCool and Flex are answering that call with a new class of turnkey racks built for the AI era. They reduce integration friction, accelerate timelines, and deliver reliable performance at scale.

In the AI age, the rack isn’t just a container for IT hardware, it’s the foundation for your compute strategy. Ready to get started? Contact us today or learn about how we can help as you migrate from air to liquid cooling.