Overcoming the “Big Commitment”: How Co-Development Will De-Risk the Future of AI Infrastructure

Key Points

• Traditional air cooling has reached its physical limits due to unprecedented AI GPU density, making advanced liquid cooling a necessity.
• Enterprise adoption is often stalled by the "big commitment" objection, fears of high upfront capital expenditure, catastrophic fluid leaks, and complex retrofits.
• A strategic co-development model across hardware OEMs, fluid suppliers and infrastructure partners establishes standardized, compatible solutions that de-risk enterprise AI investments.

AI demand is scaling so quickly that traditional data center cooling infrastructure is struggling to keep up. With high-performance GPUs and CPUs growing more powerful and packed tightly side by side to minimize signal latency, heat has become an infrastructure bottleneck. Traditional air-cooled heatsinks now occupy too much vertical rack space, decreasing overall density, while struggling to disperse the heat generated by next-generation AI processors.

With the release of NVIDIA's next-generation Blackwell and Rubin GPUs, the industry has reached an inflection point. NVIDIA's decision to require liquid cooling for these systems proves that liquid cooling is no longer optional; it is a critical part of infrastructure. As a result, the industry conversation has fundamentally changed, from whether to adopt liquid cooling to how quickly organizations can operationalize it.

However, despite the clear need for liquid cooling, many enterprise IT leaders continue to be hesitant about adopting it. They are stuck on the "big commitment" objection: a combination of upfront sticker shock, fear of the unknown, and concerns over disrupting legacy operations. To move the industry forward, cooling providers must abandon siloed innovation and embrace a co-development model that integrates technology directly with the broader ecosystem, turning perceived risks into shared strengths.

The "Big Commitment" Bottleneck

Capital expenditure (CapEx) for new cooling technology can be high, and many IT leaders hesitate to modernize existing air-cooled data centers because they expect the transition to be complex and expensive. As a result, many delay large-scale infrastructure investments in favor of extending the life of existing facilities. To meet growing AI compute demand while new data centers are under construction, hyperscalers are increasingly retrofitting existing facilities. Although retrofits can provide short-term relief, they have practical limits and often require more costly infrastructure upgrades to accommodate the newest generation of AI hardware.

Misperceptions about the risks and operating costs of liquid cooling continue to slow adoption. Although data center operators worry about catastrophic leaks and the need to retrain workers, high-performance computing (HPC) facilities have run liquid-cooled systems for years without issue. Pioneered by the Department of Energy’s National Labs, warm-water liquid cooling is now a proven standard for extreme workloads. For example, Oak Ridge National Laboratory’s exascale Frontier supercomputer uses warm-water cooling to reject massive heat loads and capture waste energy, building on prior systems where utilizing supply water above 70°F cut cooling costs by more than half.

Historically, the cooling industry has been highly fragmented, presenting customers with an array of different coolant types, coolant distribution units, and varying connector standards. Enterprises have been anxious about making a 10- to 20-year capital expenditure on unstandardized technology or being locked into single-source vendors. Fortunately, the industry is gradually developing reliable standards, such as PG25 heat transfer fluids, which significantly mitigate the risks of single-source vendors.

Solving Fragmentation Through Co-Development

The solution to ecosystem fragmentation is adopting the co-development model: the practice of designing, testing, and validating infrastructure solutions collaboratively across hardware, cooling, fluid, and data center ecosystems to reduce risk and support compatibility. Specialized cooling providers can no longer operate in a vacuum; they must collaborate directly with major original equipment manufacturers (OEMs) and silicon designers.

NVIDIA has already established simultaneous reference architecture partnerships with cooling and infrastructure companies, including Vertiv, nVent, and CoolIT. By intentionally designing and manufacturing cooling solutions alongside hardware partners, cooling solutions are engineered to seamlessly integrate with the newest high-power CPUs and GPUs, rather than relying on aftermarket retrofits.

OEMs are now designing systems around liquid cooling from the start because retrofitting becomes significantly harder at higher rack densities. Dell, HPE, and Lenovo have already integrated direct liquid cooling into their server product lines. Additionally, designing systems that easily connect to standard data center technology cooling-system rack manifolds drastically reduces the complexity of adapting existing facilities.

Mitigating Supply Chain and Operational Risks

To overcome the fear of vendor lock-in, a true co-development model purposely avoids proprietary fluid manufacturing. Instead, by collaborating with a range of global fluid suppliers to validate pharmaceutical-grade, non-conductive fluids, OEMs can ensure supply chain resilience and guarantee that the fluids are safe for operators and the environment.

The industry is actively promoting standardization to avoid fragmented ecosystems. The Open Compute Project (OCP) runs dedicated Cold Plate and Immersion sub-projects to develop open specifications, standardized interfaces, and open-source Total Cost of Ownership (TCO) calculators, allowing hesitant customers to evaluate data objectively. In October 2025, OCP formalized this through a new alliance with ASHRAE, specifically focused on data center liquid-cooling standards and best practices.

At the operational level, data center operators cannot afford workflows that require specialized skills or extensive retraining. That's why successful liquid-cooling solutions are designed to fit seamlessly into existing operational facilities. Achieving that level of simplicity requires consideration of the people who will install, operate and maintain the equipment, not just the technology itself. By involving industrial designers alongside mechanical engineers at the start of the liquid cooling design, manufacturers can create systems that feel familiar to technicians accustomed to traditional hardware. Features such as sealed chassis and slide-out rails eliminate the need to drain fluids during routine maintenance, allowing operators to service liquid-cooled servers much like air-cooled systems.

The same principle of simplicity extends beyond maintenance to deployment. As enterprises, hyperscalers and colocation providers race to add AI capacity, many are turning to prefabricated modular data centers that can be deployed much faster than traditional facilities. The modular data center market is projected to reach $67.5 billion by 2030, underscoring the industry's demand for infrastructure that can be deployed quickly without introducing unnecessary operational complexity.

Removing Barriers to Adoption Will Build the Confidence to Scale

The best way to overcome the "big commitment" objection among data center customers is for the liquid cooling industry to embrace shared validation. When cooling technology is co-developed and validated by global supercomputing centers, major OEMs, and hyperscalers, organizations feel more confident in making a long-term infrastructure investment.

This shared validation is essential as investment in AI infrastructure continues to accelerate; becoming critical on a national scale. This is reflected in the scale of AI infrastructure investment with the four largest hyperscalers expected to spend nearly $690 billion on AI infrastructure by 2026. As public and private investment continues to grow, the industry must give data center operators the confidence to invest in liquid-cooling technologies that can support the next generation of AI infrastructure.

Through co-development and standardized, de-risked frameworks, the industry can partner with data center operators to make the next generation of sustainable infrastructure a reality.

‍