The acoustic profile of a modern data center is shifting. It is no longer just the hum of fans; it is the roar of density. As computational power per square foot escalates, the mechanical systems responsible for rejecting that heat have become the primary constraint on expansion. For years, the industry relied on air as the default medium for thermal management. That era is effectively over for high-performance computing and artificial intelligence workloads. The transition is not merely a technical upgrade; it is a fundamental re-evaluation of how infrastructure is built, financed, and operated.
Speaking at Mining Disrupt Fort Lauderdale, recorded March 25–27, 2025, Shane of Data Airflow laid out what this transition demands. Data Airflow, a firm with deep roots in commercial HVAC and Bitcoin mining infrastructure, has turned its focus toward the high-density demands of AI and High-Performance Computing. Shane’s argument is direct: the current generation of data centers is hitting a wall where air cooling is no longer viable. The choice of cooling equipment now dictates the long-term profitability of the entire facility, forcing operators to look beyond initial capital costs and toward lifecycle efficiency.
The Air Limit: Why Density Breaks Traditional Cooling
The most immediate impact of higher density is the energy penalty of cooling. Shane noted that in many modern facilities, the cooling system itself consumes a massive portion of the total power budget. This is not a minor inefficiency; it is a structural flaw in older designs that assumed lower heat loads per square foot. As GPU and ASIC densities increase, the heat generated in a confined space exceeds the capacity of air to absorb and dissipate it quickly enough. The result is a forced migration away from standard air-cooled units toward hydronic solutions, including direct-to-chip cooling, rear-door heat exchangers, and immersion systems.
we are truly hitting a cooling crisis in the US.
Shane · 1:56
The reasoning is thermodynamic: air simply does not have the thermal mass or density to handle the heat flux of next-generation compute. For operators, this means a significant capital expenditure to re-engineer their cooling plants. However, it also unlocks higher power density per rack, allowing more compute to be packed into the same physical footprint. The hidden costs of inefficient equipment—energy waste, downtime, and equipment damage—become the primary drivers of operational loss, making the right equipment selection a matter of financial survival rather than just technical preference.
The Water Dilemma: Dry Cooling Where Water Is Scarce
However, the shift to liquid cooling introduces a new constraint: water. In many parts of the United States, water allocation is tightening, and regulatory pressure is mounting to reduce consumption. This creates a conflict between the desire for high-efficiency evaporative cooling, which relies heavily on water, and the need for operational resilience in water-scarce regions. Shane highlighted the rise of dry cooling systems, specifically large dry coolers, as a critical component of modern design. These systems reject heat without utilizing water, offering a path to operational independence from municipal water supplies.
The trade-off is not without cost. Dry cooling is generally less efficient than evaporative cooling in warm climates, meaning it may require more energy to achieve the same temperature reduction. Yet, for sites where water is unavailable or too expensive, it is the only viable option. Data Airflow’s approach involves selecting equipment that balances these factors, often using a hybrid system that can switch between modes based on ambient conditions and water availability. The goal is to maintain equipment lifespan and operational uptime without incurring the penalty of water usage or the inefficiency of pure dry cooling in peak heat.
Sizing for Resilience: The Economics of Uptime
Beyond the physics of heat transfer, the economics of uptime drive cooling design. In the data center industry, uptime is not a binary state; it is a tiered specification that dictates the level of redundancy and maintenance required. Shane emphasized that for publicly traded companies or those with significant investor stakes, the promise of uptime is a financial covenant. A Tier 3 data center, for example, allows for only 1.6 hours of downtime per year. This stringent requirement forces operators to size their cooling systems not just for peak load, but for resilience against failure.
a lot of people don't know that chillers run most efficiently at 70 to 80% run rate
Shane · 3:57
This efficiency sweet spot has profound implications for system design. Running chillers at 100% capacity continuously is not only less efficient but also accelerates wear and tear on the equipment. By designing for a 70 to 80% run rate, operators can extend the life of their assets and reduce maintenance costs. This requires a more sophisticated approach to sizing, where the total cooling capacity is slightly higher than the peak load, allowing the system to operate in its most efficient band for the majority of the year. It is a strategy that prioritizes long-term operational cost over initial capital expenditure.
The Deployment Race: Navigating Power Grid Delays
The urgency of these design choices is compounded by the bottleneck in power availability. In many US markets, the wait time for new grid connections is stretching into years, not months. Shane cited a local market where the utility company estimated a seven-year wait for new power capacity. This disparity between the rapid pace of AI adoption and the slow pace of grid expansion means that operators must be extremely precise in their planning. There is no room for error; if the cooling system is not perfectly sized and commissioned, the entire investment is at risk.
the local municipality says if you want real power, it's going to take them seven years
Shane · 14:37
This delay forces a re-evaluation of deployment strategies. Operators are looking for ways to accelerate the installation process, and Shane said Data Airflow keeps contracts and partnerships in the industry to get equipment on site quickly, ensuring that when power becomes available, the facility can be brought online rapidly. The focus is on minimizing the time between power connection and full operational capacity, a period where every day of delay represents lost revenue.
Free Cooling: Engineering the Off-Peak Advantage
One of the most effective ways to offset the high energy costs of cooling is to exploit natural ambient conditions. Shane described a technique known as free cooling, where the system bypasses the energy-intensive chiller and uses external air to cool the fluid. This is achieved by installing a heat exchanger in parallel with the chiller. When the outside temperature drops below a certain threshold, the system automatically switches to this mode, using only fans and pumps to circulate the fluid.
when you get below a certain ambient temperature, we turn off the chiller
Shane · 18:33
The benefits of free cooling are substantial. In many climates, a significant portion of the year falls within the temperature range where this mode can operate. By eliminating the need for refrigerant compression during these hours, operators can dramatically reduce their power usage effectiveness (PUE). This lowers operational costs. For investors, this translates to a more predictable and lower cost structure, making the project more attractive in a competitive market.
The session concluded with a look at the broader market context. Shane noted that data center expansion is expected to double in the next five years, a goal he described as aggressive but aligned with the trajectory of AI and HPC. The Bitcoin price on March 25, 2025, was $87,393, while by October 2026, it had adjusted to $82,811. This modest decline reflects broader market volatility but does not change the technical imperative. The shift to liquid cooling is not a temporary trend but a permanent structural change driven by the physics of increasing density. The challenges of water scarcity, power grid delays, and the economic imperative for efficiency remain as pressing as they were in 2025. As the industry looks forward, the fundamental constraints identified by Data Airflow will remain the central focus. The operators who plan cooling, water and power together will be the ones who get to build.