AWS Liquid Cooling Strategy 2025: How In-House Tech and NVIDIA Partnership Reshape Data Center Energy

Industry Adoption: How AWS Evolved from Liquid Cooling Necessity to Market Disruption

Between 2021 and 2024, Amazon Web Services (AWS) treated liquid cooling as a strategic necessity to prepare for the future. The primary driver was the undeniable trajectory of AI chip power consumption, which was on course to exceed 1,000W per chip, rendering traditional air cooling obsolete. This period was characterized by foundational moves: securing massive, carbon-free power through the $650 million acquisition of Talen Energy’s nuclear-powered data center campus, forming forward-looking R&D partnerships with firms like Orbital Materials to explore next-generation materials, and announcing a landmark collaboration with NVIDIA in late 2023. This NVIDIA partnership was the key inflection point, designating the upcoming GH200 instances as the first on AWS to feature liquid cooling. The culmination of this preparatory phase was the December 2024 announcement of a new hybrid, retrofittable cooling system, signaling a clear shift from planning to productization.

The period from January 2025 to the present marks a dramatic acceleration from preparation to market disruption and scaled execution. AWS is no longer just an adopter of liquid cooling; it has become a vertically integrated innovator. The centerpiece of this shift is the proprietary In-Row Heat Exchanger (IRHX), a custom direct-to-chip system co-engineered with NVIDIA in just 11 months. This move directly challenges third-party vendors like Vertiv, whose stock fell over 6% after the announcement, and establishes AWS as a hardware developer. The strategy was immediately commercialized with the launch of the P6e-GB200 UltraServers, representing the first large-scale deployment of this custom technology. This technical capability has become a powerful commercial tool, enabling massive contracts like the $38 billion multi-year agreement with OpenAI, which relies on the ability to cool immense clusters of NVIDIA’s latest GPUs. The variety of activities—from in-house hardware design and ODM partnerships with Quanta to massive customer contracts and ambitious sustainability targets—demonstrates that liquid cooling has evolved from a technical solution into a core pillar of AWS’s competitive strategy, enabling performance, efficiency, and market capture.

Table: AWS Strategic Investments in AI Infrastructure and Cooling

Table: AWS Liquid Cooling Partnership Ecosystem

Geography: AWS Expands Its Liquid-Cooled Footprint from US Foundations to Global AI Hubs

Between 2021 and 2024, AWS’s geographic focus for next-generation data center infrastructure was predominantly domestic and foundational. The key event was the March 2024 acquisition of a data center campus in Pennsylvania, notable for its direct connection to a nuclear power plant. This move was not about immediate liquid cooling deployment but about securing a massive, stable, carbon-free power source in the U.S. to support future, energy-intensive AI workloads that would inevitably require advanced cooling. The strategy was centered on preparing the U.S. power and real estate backbone for the coming AI-driven energy demand.

From 2025 onward, the strategy has rapidly globalized and shifted towards active deployment. While the U.S. remains a critical hub, evidenced by the $11 billion investment in Georgia for AI infrastructure and a nationwide initiative to use recycled wastewater in 120 U.S. data centers, international expansion is now a primary focus. The June 2025 partnership with SK Group to build an AI Zone in Ulsan, South Korea, is a landmark move, establishing a major liquid-cooled AI hub in Asia. Simultaneously, the partnership with GE Vernova targets data center decarbonization across North America, Europe, and Asia. This geographic diversification shows that AWS is now exporting its liquid cooling blueprint globally. The primary driver is the need to deploy high-performance AI services closer to international customers while navigating regional energy and water constraints, a risk AWS is proactively addressing through its water recycling and efficiency initiatives.

Technology Maturity: AWS’s Journey from Pilot Planning to Scaled, In-House Production

In the 2021–2024 period, liquid cooling at AWS was in a pre-commercial, strategic planning phase. The technology was being evaluated and designed for a future that had not yet arrived. The most significant validation point was the November 2023 announcement that upcoming AWS instances with NVIDIA GH200 GPUs would be its *first* to be liquid-cooled, signaling the transition from R&D to a concrete product roadmap. The development of a hybrid, modular cooling system, unveiled in December 2024, represented the productization of this R&D. It was designed to be configurable and retrofittable, but at this stage, it was still a technology preparing for deployment rather than one operating at scale. The focus was on ensuring a solution would be ready for chips projected to exceed 1,000W.

The period from 2025 to today represents a significant leap in maturity to *scaled commercial deployment* and *vertical integration*. The technology is no longer a planned feature but a deployed product and a competitive advantage. The launch of the proprietary In-Row Heat Exchanger (IRHX), developed in-house with NVIDIA, marks a shift from being a technology consumer to a technology creator. Its immediate integration into the commercially available P6e-GB200 UltraServers proves its readiness for production workloads. Furthermore, the engagement of Quanta as an ODM partner to manufacture and integrate the IRHX system confirms the move from bespoke solutions to mass production. AWS’s publicly stated goal to have over 20% of its machine learning capacity liquid-cooled by 2026 is the ultimate validation point, demonstrating that the technology has moved far beyond pilots and is now a core component of its scaled infrastructure strategy.

Table: SWOT Analysis of AWS’s Liquid Cooling Strategy

Forward-Looking Insights and Summary

The data from 2025 reveals that AWS’s liquid cooling strategy has reached an aggressive execution phase, transforming the company from a technology consumer into a market-shaping producer. The year ahead will be defined by the rapid deployment of its proprietary IRHX system, moving beyond initial server launches to a broad-based retrofitting program across its global fleet. The most critical signal to watch will be AWS’s progress toward its goal of having over 20% of its multi-gigawatt machine learning capacity liquid-cooled by 2026. Achieving this target will validate the scalability and economic viability of its in-house approach and cement its competitive advantage in delivering high-performance AI at a lower total cost of ownership.

Market actors should pay close attention to the ripple effects on the data center supply chain. The traditional cooling vendor market, exemplified by Vertiv’s stock reaction, is under pressure, while opportunities are surging for ODMs like Quanta and component suppliers like Fositek who can align with AWS’s custom architecture. Another key signal gaining traction is the fusion of high-performance computing with aggressive sustainability goals. The recycled water initiative for 120 U.S. data centers is not a peripheral activity but a critical enabler of its liquid cooling strategy, mitigating operational risk in water-scarce regions. Expect AWS to further integrate its custom cooling technology with its energy procurement and water management strategies, creating a holistic efficiency model that competitors will be forced to emulate. The era of simply buying off-the-shelf cooling is losing steam for hyperscalers; the future is custom, integrated, and scaled.

Frequently Asked Questions

Why did AWS suddenly shift to developing its own liquid cooling technology?
The shift was driven by the increasing power consumption of AI chips, which are projected to exceed 1,000W and make traditional air cooling ineffective. By developing its proprietary In-Row Heat Exchanger (IRHX) system with NVIDIA, AWS moved from being a technology adopter to a vertically integrated innovator. This allows them to create more efficient, powerful, and cost-effective AI infrastructure, giving them a competitive edge to win massive contracts like the $38 billion deal with OpenAI.

What is the In-Row Heat Exchanger (IRHX) and why is it significant?
The IRHX is a proprietary, direct-to-chip liquid cooling system that AWS co-engineered with NVIDIA in just 11 months. Its significance is twofold: first, it marks AWS’s entry into hardware development, directly challenging traditional third-party vendors like Vertiv. Second, it is a key technical enabler, allowing AWS to efficiently cool the latest high-density NVIDIA GPUs (like the Blackwell series), which is essential for large-scale AI deployments such as their P6e-GB200 UltraServers.

How is AWS’s partnership with NVIDIA central to its liquid cooling strategy?
The NVIDIA partnership was the key catalyst. It began in late 2023 with the announcement that NVIDIA’s GH200 instances would be the first on AWS to use liquid cooling. This collaboration then deepened significantly, leading to the co-engineering of AWS’s proprietary IRHX system specifically for NVIDIA’s Blackwell GPUs. This deep integration allows AWS to offer optimized performance and efficiency for the most sought-after AI hardware on the market.

Isn’t using liquid cooling bad for the environment due to high water consumption?
AWS is proactively addressing this concern. The article highlights that a core part of their strategy is sustainability. To mitigate water usage, AWS announced a major initiative in June 2025 to use recycled wastewater for cooling at 120 of its U.S. data centers by 2030. This, combined with its broader goal to be water-positive through water replenishment projects, shows that AWS is integrating water management directly into its cooling strategy to reduce environmental impact and operational risk.

What does this strategy mean for other companies in the data center market?
AWS’s move creates both challenges and opportunities. Traditional, off-the-shelf cooling vendors like Vertiv face a significant threat as hyperscalers bring technology in-house. Conversely, the strategy creates new opportunities for Original Design Manufacturers (ODMs) like Quanta, which is building the server racks with the integrated IRHX system, and specialized component suppliers like Fositek, which is mass-producing parts for AWS’s custom systems. The market is shifting to reward partners who can align with and support these custom, vertically integrated architectures.