Liquid Cooling to Define Next AI Infrastructure Wave as Grid Strains Under Compute Demand

The AI boom has moved past the model-building phase. The exponential growth curve is now hitting the physical world, forcing a reckoning with energy and cooling. The investment paradigm is shifting from software and algorithms to the fundamental infrastructure required to run them-a new layer of the adoption curve where the "plumbing" determines the pace of the entire next paradigm.

The scale of this buildout is staggering. U.S. data-center spending alone is expected to exceed half a trillion dollars in 2025. This isn't just about adding servers; it's about constructing the entire energy and cooling ecosystem at an unprecedented rate. Yet this rapid expansion is creating a severe bottleneck. Data centers now consume about 4.4% of U.S. electricity, a figure that is projected to climb to as high as 12.0% by 2028. In some regions, AI-driven demand is already outpacing available grid capacity, forcing companies to delay projects or install inefficient backup generators.

This is the core thesis: the "plumbing" for AI is not just software pipelines. It is the entire energy and cooling infrastructure that must scale exponentially to support the next paradigm. The invisible bottleneck is no longer just about compute power; it's about the physical systems that manage the immense heat and power these systems generate. As one expert notes, the question that determines success is often the most boring one: Do we have the plumbing to actually run this thing?. The transition from pilot to production fails when this infrastructure is ignored. The exponential growth of AI is now constrained by the linear limits of energy grids and cooling capacity, making this physical layer the critical next frontier for investment and innovation.

The Cooling Conundrum: A Market in Exponential Growth

The bottleneck is now a specific technology. As AI workloads push server densities to new extremes, air cooling is hitting its physical limits. The market is responding with a clear, exponential shift toward liquid cooling, a move that defines the next phase of the infrastructure S-curve.

The scale of this buildout is projected to be massive. The global data center cooling market is expected to reach $40–45 billion by 2030. Within that, liquid cooling alone is forecast to account for $15–20 billion. This isn't a niche upgrade; it's becoming the default standard. Data shows that more than 50% of new hyperscale capacity will be liquid-cooled by 2027. The industry consensus has solidified: liquid cooling is the most viable path forward for supporting the performance and scalability of next-generation AI.

This rapid adoption is driving a wave of consolidation. The market is maturing from a fragmented landscape of small vendors into a capital-intensive industry where scale and integration matter. Recent deals signal this shift. In late 2024, Eaton made a $9.5 billion acquisition of Boyd Thermal, a major move to accumulate liquid cooling expertise. This trend continued throughout 2025 with other significant purchases, including Trane's buy of Stellar Energy Digital and Daikin's acquisition of Chilldyne. The goal is clear: to build comprehensive portfolios that can deliver the high-capacity, reliable cooling solutions required for AI's exponential growth.

The technological race within liquid cooling is also accelerating. While immersion cooling-submerging entire servers in fluid-remains a prominent strategy, it introduces operational trade-offs. This has spurred innovation in targeted approaches like LiquidRack™, a server-level spray liquid cooling solution designed to offer high thermal performance with easier serviceability. At the same time, open standards are emerging, with Google's Project Deschutes specifications now available to the Open Compute Project, aiming to drive interoperability and accelerate deployment. The bottom line is that the cooling layer is no longer an afterthought. It is a critical, high-growth infrastructure segment where technological leadership and capital scale will determine which companies ride the next leg of the AI adoption curve.

Energy Nexus: The Critical Bottleneck and Investment Implications

The energy bottleneck is the most visible choke point on the AI adoption curve. As data center demand surges, it is directly outpacing local energy capacity. This isn't a theoretical future risk; it's a present-day constraint forcing companies to take drastic measures. In regions where AI-driven demand is highest, operators are delaying projects, contracting power directly from private producers, and installing multiple, inefficient reciprocating generators. The physical reality is that the grid simply cannot keep up with the exponential buildout of compute.

This creates a dual investment thesis. First, there is the established path of providing the physical cooling infrastructure that enables this power-hungry hardware to function. The second, and equally critical, frontier is enabling flexible, grid-integrated power solutions. The system is under strain, and the solution requires more than just building more power plants. It demands smarter integration-using battery storage, demand-response systems, and distributed generation to manage the massive, variable load from AI workloads. The companies that succeed will be those that can deliver both the high-capacity cooling and the intelligent power management needed to keep the lights on.

The systemic risk here is profound. The energy nexus-the interplay between AI's demand for power, water, and critical minerals-is a choke point that, if mismanaged, could choke AI's growth itself. The recent history is a warning. In July 2024, a voltage fluctuation in northern Virginia triggered the simultaneous disconnection of 60 data centers, creating a 1,500-megawatt surplus that forced emergency grid adjustments. This incident underscores how data centers can impact grid reliability, turning a local energy shortage into a regional risk. Failure to manage this nexus risks not just project delays, but broader grid instability, rising consumer costs, and a loss of the social license for AI to expand.

The bottom line is that early infrastructure investment is a strategic necessity, not a luxury. The exponential growth of AI is now constrained by the linear limits of energy grids and cooling capacity. The companies building the fundamental rails for this next paradigm-those providing the physical cooling infrastructure and those enabling flexible, grid-integrated power solutions-are positioned at the critical intersection of this S-curve. Their work is the essential plumbing that will determine whether the AI boom can sustain its exponential pace or hits a hard wall.

Catalysts and Risks: What to Watch in the Infrastructure Buildout

The infrastructure buildout is now in motion, but its pace will be dictated by a mix of regulatory catalysts and physical constraints. The next phase of the AI S-curve will be shaped by decisions on siting and power, as well as the tangible mismatch between cooling capacity and compute demand.

Regulatory catalysts are emerging as a key near-term driver. The boom in hyperscale data center development is already outpacing available capacity in some regions, forcing operators to take extreme measures like installing inefficient backup generators. This strain is prompting policy shifts. In major markets, debates are intensifying over reliability and affordability, with legislative actions like Texas Senate Bill 6 signaling a move toward greater market intervention. The critical test will be how regulators balance the economic incentives that have fueled the buildout-such as discounted energy tariffs-with the need to ensure grid stability. A major grid incident, like the voltage fluctuation in northern Virginia in July 2024 that disconnected 60 data centers, serves as a stark warning. Such events are likely to accelerate regulatory scrutiny and could force changes in how data centers are sited and powered.

The primary risk is a tangible mismatch between cooling capacity and AI compute demand. While the market is consolidating and scaling liquid cooling solutions, the physical deployment of these systems cannot keep pace with the exponential ramp-up of AI hardware. This creates a direct risk of project delays and higher costs for operators who must wait for cooling infrastructure or resort to less efficient alternatives. The industry's push toward higher-capacity cooling units, like those capable of managing 2MW of processing power, is a response to this pressure. Yet, if the cooling layer lags, it will bottleneck the entire compute stack, derailing the adoption curve at a critical juncture.

On the positive side, the expansion of pre-existing data center clusters is a powerful, self-reinforcing catalyst. Northern Virginia, for instance, is the top market in the U.S. and is likely to retain its leading status as the ecosystem of power, fiber, and skilled labor matures. These established hubs offer a proven infrastructure layer that reduces the regulatory and logistical friction for new projects. The continued expansion within these clusters provides a more predictable and lower-risk path for the next wave of investment, acting as a foundation for the broader buildout.

The bottom line is that the infrastructure S-curve is now navigating a complex landscape of policy and physical constraints. Success will favor those who can navigate regulatory shifts, scale cooling solutions to match compute demand, and leverage the advantages of established clusters. The risks are real, but the catalysts are also material, setting the stage for a decisive period in the buildout.