The Data Center Cooling Revolution - Part 1: Why It Matters for a Sustainable Future
Welcome to part one of our two-part series on the future of data center cooling. In this series, we’ll explore how the explosive growth of AI and digital infrastructure is driving urgent innovation in cooling technologies, and why this matters for planetary health.
The AI Boom and the Cooling Challenge
Artificial intelligence, cloud computing, and digital transformation are reshaping industries at an unprecedented pace. By 2030, data centers are projected to consume nearly 1,000 TWh of electricity (roughly equal to the annual electricity use of Canada and South Korea combined) with more than 80% of that demand coming from the US and China. While this represents just 1.5% of global electricity use, the rapid growth of data centers threatens to strain electric grids worldwide.
A key driver of this demand is cooling: approximately 40% of a data center’s energy use is devoted to keeping servers and chips from overheating. As chips and servers work harder, they generate more heat, creating a cyclical challenge where more energy is needed just to remove the heat produced by computing.
If chips exceed their optimal temperature, they “throttle”, slowing down to prevent damage. Data center operators face a stark choice: make computing more efficient to reduce heat generation, or develop better cooling systems that are faster and less energy-intensive.
Cooling Technologies: From Air to Liquid Cooling
Historically, data centers have relied on air cooling at the room level, using basic fans paired with hot and cold aisle layouts to manage airflow. This method, while effective for lower energy density racks, is increasingly inadequate as computational power rises. Today’s most advanced chips can exceed 1,000 W at the chip level, with some even approaching 2,700 W, far beyond the capabilities of traditional air cooling. With many processors making up a single rack, rack densities have entered the kW range.
This has led to the rise of liquid cooling technologies, which use water or refrigerants to extract heat more efficiently. For chips above 200 W, liquid cooling becomes essential. Direct-to-chip (D2C) cooling brings coolant directly to the chip, allowing for precise temperature control and significant reductions in energy and water use. For the most power-dense racks, immersion cooling, where entire servers are submerged in dielectric fluids, offers the ultimate in thermal management.
Power Usage Effectiveness (PUE): The Efficiency Metric
Data center efficiency is measured by Power Usage Effectiveness (PUE), the ratio of total facility energy use to the energy used by computing equipment. A perfect PUE is 1.0, meaning all energy consumed goes to computing. Today, top-tier data centers operate at PUEs of 1.2–1.5, but industry-wide progress in further lowering PUE has slowed in recent years. Achieving further gains will require new technologies and approaches.
The Water Factor
While many liquid cooling systems operate in closed loops (minimizing water use), most data centers still rely on water-cooling towers, especially in hot, arid regions. These towers evaporate water to dissipate heat, increasing both energy and water consumption. Reducing or eliminating cooling towers can have a significant impact on both electricity and water use, aligning with broader sustainability goals.
Why This Matters
As digital infrastructure expands to support AI, IoT, and cloud computing, efficient and sustainable cooling is becoming both a critical challenge and a unique opportunity to invest in breakthrough hard tech that can drive both environmental and economic benefits.
Stay tuned for Part 2, where we’ll dive deeper into the market landscape, highlight leading innovations in on-chip cooling, and share our perspective on the venture opportunities ahead.