Vertiv outlines five key trends reshaping AI data centres
Vertiv has identified higher voltage power architectures, digital twin modelling and adaptive liquid cooling as central trends that it expects will shape data centre design and operations in the coming years as AI workloads expand.
The company has published a new Vertiv Frontiers report that outlines how AI and high-performance computing are changing data centre infrastructure. The report links these changes to four macro forces: extreme densification, gigawatt-scale deployments, the treatment of the data centre as a single unit of compute, and diversification in processor types.
Vertiv said these forces are driving five main trends across power, cooling, energy strategy and facility design. The trends focus on support for AI training and inference workloads in both hyperscale and enterprise environments.
"The data centre industry is continuing to rapidly evolve how it designs, builds, operates and services data centres, in response to the density and speed of deployment demands of AI factories," said Vertiv chief product and technology officer, Scott Armul, Vertiv. "We see cross-technology forces, including extreme densification, driving transformative trends such as higher voltage DC power architectures and advanced liquid cooling that are important to deliver the gigawatt scaling that is critical for AI innovation. On-site energy generation and digital twin technology are also expected to help to advance the scale and speed of AI adoption."
Power shift
The first trend identified in the report is a shift in power distribution. Vertiv said most current data centres rely on hybrid AC/DC power distribution from the grid to IT racks. This approach typically involves three to four power conversion stages and associated inefficiencies.
Rising rack power densities from AI workloads are putting this model under strain. Vertiv expects a move towards higher voltage DC architectures in response. The company said this change reduces current, reduces the size of conductors and cuts the number of conversion stages. It also centralises power conversion at the room level.
Hybrid AC and DC systems remain common in current facilities. Vertiv expects higher voltage DC systems to become more prevalent as rack densities increase and as DC standards and equipment mature. The company also links on-site generation and microgrids with the adoption of higher voltage DC.
Distributed AI
The second trend focuses on where AI inference will run. Vertiv said that large investments in AI data centres so far have supported widespread use of large language models by consumers and businesses. The company expects AI to become more central to business operations.
The report states that the location of inference services will vary by organisation. Factors will include security, data residency and latency. Vertiv said industries such as finance, defence and healthcare may require private or hybrid AI environments. These would run via on-premise data centres rather than solely through public cloud facilities.
Vertiv said operators can increase AI capacity through new builds or retrofits of existing sites. It flagged scalable high-density power infrastructure and liquid cooling systems as enablers for these changes.
Energy autonomy
The third trend addresses energy strategy. Vertiv said data centres have relied on short-term on-site generation for many years. Traditional arrangements support resiliency rather than ongoing operation.
According to the report, wider grid constraints are now encouraging extended energy autonomy, particularly for AI-focused data centres. Vertiv pointed to investment in on-site power generation using natural gas turbines and other technologies. The company said power availability challenges are the primary driver.
It also highlighted technology strategies it describes as "Bring Your Own Power (and Cooling)". Vertiv expects such approaches to form part of long-term energy autonomy plans for operators that face capacity constraints or grid connection delays.
Digital twins
The fourth trend centres on digital twin-driven design and operations. Vertiv said increasingly dense AI workloads and more powerful GPUs are pushing operators to deploy new facilities at higher speed.
The company said operators can use AI-based tools to create virtual models of data centres. These digital twins map and specify both IT systems and critical digital infrastructure. Vertiv said such designs are often linked with prefabricated modular components.
These components can be deployed as standardised units of compute. Vertiv said this approach can reduce what it calls "time-to-token" by up to 50 per cent. It described gigawatt-scale buildouts as a future requirement for AI expansion and stated that digital twins will be important for such programmes.
Liquid cooling
The fifth trend in the report focuses on liquid cooling. Vertiv said AI workloads and supporting infrastructure have increased adoption of liquid cooling among data centre operators. It said liquid systems are now mission-critical for a growing number of sites.
The company also highlighted the role of AI in improving cooling. It said AI, combined with more extensive monitoring and control systems, can refine and optimise liquid cooling set-ups. This combination could predict potential failures in advance. It could also manage fluid and components more effectively.
Vertiv said this trend should lead to higher reliability and uptime for high-value hardware and associated data and workloads. The company expects the intersection of AI and liquid cooling design to remain an area of active development.
Vertiv operates in more than 130 countries and serves data centres, communication networks and commercial and industrial facilities. The company develops power management, thermal management and IT infrastructure products and services for use from large cloud sites through to edge locations.
"On-site energy generation and digital twin technology are also expected to help to advance the scale and speed of AI adoption," said Armul.