Innovation in powering and cooling AI racks, management of energy consumption and emissions all to be a focus in new year
Vertiv experts anticipate increased industry innovation and integration to support high-density computing, regulatory scrutiny around AI, as well as increasing focus on sustainability and cybersecurity efforts.
“Our experts correctly identified the proliferation of AI and the need to transition to more complex liquid- and air-cooling strategies as a trend for 2024, and activity on that front is expected to further accelerate and evolve in 2025,” said Vertiv CEO Giordano (Gio) Albertazzi. “With AI driving rack densities into three- and four-digit kWs, the need for advanced and scalable solutions to power and cool those racks, minimize their environmental footprint, and empower these emerging AI Factories has never been higher. We anticipate significant progress on that front in 2025, and our customers demand it.”
Shrirang Deshpande, Country Head – AI Practice and Strategic Programs, Vertiv “This year witnessed the AI revolution taking center stage, with the momentum anticipated to grow in the coming year and beyond. In India, our government is also playing a pivotal role in accelerating AI adoption and supporting the development of a sovereign AI Cloud. With our comprehensive power and cooling portfolio, including the new Vertiv™ 360AI portfolio, we are well-positioned to address the dynamic demands of customers exploring AI support. Combined with a global service network and expert engineers, we’re ready to support transformative AI journeys”.
The 2025 trends most likely to emerge across the data center industry, according to Vertiv experts:
- Power and cooling infrastructure innovates to keep pace with computing densification: In 2025, the impact of compute-intense workloads will intensify, with the industry managing the sudden change in a variety of ways. Advanced computing will continue to shift from CPU to GPU to leverage the latter’s parallel computing power and the higher thermal design point of modern chips. This will further stress existing power and cooling systems and push data center operators toward cold-plate and immersion cooling solutions that remove heat at the rack level. Enterprise data centers will be impacted by this trend, as AI use expands beyond early cloud and colocation providers.
- AI racks will require UPS systems, batteries, power distribution equipment and switchgear with higher power densities to handle AI loads that can fluctuate from a 10% idle to a 150% overload in a flash.
- Hybrid cooling systems, with liquid-to-liquid, liquid-to-air and liquid-to-refrigerant configurations, will evolve in rackmount, perimeter and row-based cabinet models that can be deployed in brown/greenfield applications.
- Liquid cooling systems will increasingly be paired with their own dedicated, high-density UPS systems to provide continuous operation.
- Servers will increasingly be integrated with the infrastructure needed to support them, including factory-integrated liquid cooling, ultimately making manufacturing and assembly more efficient, deployment faster, equipment footprint smaller, and increasing system energy efficiency.
- Data centers prioritize energy availability challenges: Overextended grids and skyrocketing power demands are changing how data centers consume power.Globally, data centers use an average of 1-2% of the world’s power, but AI is driving increases in consumption that are likely to push that to 3-4% by 2030. Expected increases may place demands on the grid that many utilities can’t handle, attracting regulatory attention from governments around the globe – including potential restrictions on data center builds and energy use – and spiking costs and carbon emissions that data center organizations are racing to control. These pressures are forcing organizations to prioritize energy efficiency and sustainability even more than they have in the past.
In 2024, we predicted a trend toward energy alternatives and microgrid deployments, and in 2025 we are seeing an acceleration of this trend, with real movement toward prioritizing and seeking out energy-efficient solutions and energy alternatives that are new to this arena. Fuel cells and alternative battery chemistries are increasingly available for microgrid energy options. Longer-term, multiple companies are developing small modular reactors for data centers and other large power consumers, with availability expected around the end of the decade. Progress on this front bears watching in 2025.
- Industry players collaborate to drive AI Factory development: Average rack densities have been increasing steadily over the past few years, but for an industry that supported an average density of 8.2kW in 2020, the predictions of AI Factory racks of 500 to 1000kW or higher soon represent an unprecedented disruption. As a result of the rapid changes, chip developers, customers, power and cooling infrastructure manufacturers, utilities and other industry stakeholders will increasingly partner to develop and support transparent roadmaps to enable AI adoption. This collaboration extends to development tools powered by AI to speed engineering and manufacturing for standardized and customized designs. In the coming year, chip makers, infrastructure designers and customers will increasingly collaborate and move toward manufacturing partnerships that enable true integration of IT and infrastructure.
- AI makes cybersecurity harder – and easier: The increasing frequency and severity of ransomware attacks is driving a new, broader look at cybersecurity processes and the role the data center community plays in preventing such attacks. One-third of all attacks last year involved some form of ransomware or extortion, and today’s bad actors are leveraging AI tools to ramp up their assaults, cast a wider net, and deploy more sophisticated approaches. Attacks increasingly start with an AI-supported hack of control systems, embedded devices or connected hardware and infrastructure systems that are not always built to meet the same security requirements as other network components. Without proper diligence, even the most sophisticated data center can be rendered useless.
As cybercriminals continue to leverage AI to increase the frequency of attacks, cybersecurity experts, network administrators and data center operators will need to keep pace by developing their own sophisticated AI security technologies. While the fundamentals and best practices of defense in depth and extreme diligence remain the same, the shifting nature, source and frequency of attacks add nuance to modern cybersecurity efforts.
- Government and industry regulators tackle AI applications and energy use: While our 2023 predictions focused on government regulations for energy usage, in 2025, we expect the potential for regulations to increasingly address the use of AI itself. Governments and regulatory bodies around the world are racing to assess the implications of AI and develop governance for its use. The trend toward sovereign AI – a nation’s control or influence over the development, deployment and regulation of AI and regulatory frameworks aimed at governing AI – is a focus of The European Union’s Artificial Intelligence Act and China’s Cybersecurity Law (CSL) and AI Safety Governance Framework. Denmark recently inaugurated their own sovereign AI supercomputer, and many other countries have undertaken their own sovereign AI projects and legislative processes to further regulatory frameworks, an indication of the trajectory of the trend. Some form of guidance is inevitable, and restrictions are possible, if not likely.
Initial steps will be focused on applications of the technology, but as the focus on energy and water consumption and greenhouse gas emissions intensifies, regulations could extend to types of AI application and data center resource consumption. In 2025, governance will continue to be local or regional rather than global, and the consistency and stringency of enforcement will widely vary.
