PROMERSION ANALYTICS

Promersion has shared many analytical visuals, graphs, and conceptual frameworks in public presentations, industry discussions, and published articles. These materials are intended to make complex market dynamics, technology shifts, and industry trade-offs accessible and easy to interpret.

The content presented here is drawn from these published materials and is intentionally high level. These visuals reflect Promersion’s analytical approach, combining technical understanding, market structure, and practical industry context, while remaining transparent about scope and limitations. Links to the published content are provided to facilitate improved context.

Access to more detailed assets like raw data, assumptions, and analytical context is reserved under license for Promersion clients as part of tailored consulting engagements.

You are welcome to reference and reuse the materials on this page responsibly. These visuals may be shared publicly with appropriate credit. Graphs may be recreated for non-commercial purposes, provided they accurately reflect the data and context presented and clearly acknowledge Promersion as the source.

Over the past year, the data coming out of the datacenter industry has been difficult to ignore. Not announcements or marketing narratives, but numbers tied to capacity, capital allocation, power reservations, and physical build-out. When those data points are aggregated and projected forward, the outcome is impressive.

This chart reflects where that aggregation currently lands. What stands out is not just continued growth, but the absolute scale the industry appears to be moving toward within a relatively short time frame.

2025

• Microsoft, Google, Meta, AWS all deploy multi-GW AI campuses
• NVIDIA supply ramps (GB200/NVL72)
• Grid pre-allocations from 2023–2024 become physical builds
• AI-optimized datacenters (50–200 kW per rack) accelerate
• Sovereign AI projects begin to break ground
• China pushes its homegrown GPU ecosystems into mass deployment

2026

• Simultaneous global buildout
• The non-elastic demand for AI capacity
• The beginning of the liquid cooling mass inflection
• The hyperscale shift to internal platform development
• Multi-facility construction cycles across US, EU, Middle East, and APAC

2028–2030

• AI deployments shift from “land grab” to “fleet refresh”
• Liquid cooling becomes normalized across OEM platforms
• Power constraints slow physical expansion
• New regions absorb expansion at a steadier pace
• Facilities catch up with the AI-first era

2031–2035

• AI architectures mature
• Hardware refresh cycles stabilize
• Power grid expansion plateaus
• Market growth shifts to global south/satellite regions
• Efficiency improvements slow capex intensity

Thermal Design Power (TDP) for high-performance chips, particularly AI GPUs and processors, is experiencing a dramatic rise, with next-generation units projected to reach 2 000 W to over 5 000W, far exceeding the limits of traditional air cooling. This rapid increase is driven by the need for higher computing power in AI and high-performance computing (HPC) workloads and is a major driver for liquid cooling in the industry.

INTEL

AMD

NVIDIA

The IT equipment landscape in 2030 will likely be vastly different, shaped by evolving compute demands, technological innovation and increasingly complex thermal management requirements. While this projection should be treated as speculative, it offers some valuable insights into the potential trajectory for the industry.

Decline of General-Purpose Systems

General-purpose systems, once a cornerstone of the ecosystem, will likely shrink as specialized systems replace them. These systems will persist and cover a larger variation in load per system (0.8-3 kW) in use cases requiring flexibility but may no longer dominate the market.

Evolution of Cloud-Optimized Platforms

Cloud-optimized systems will likely remain significant. Although their design should adapt to handle more diverse workloads efficiently, the architecture and power load per system (0.8-1.8 kW) will likely remain relatively stable to ensure flexibility.

Transition of Modular HPC Compute

By 2030, modular HPC systems are expected to be absorbed into other categories, including Heterogeneous Compute, AI/GPU-Optimized Systems and Edge Computing. This reflects a trend toward greater specialization and the adaptation of modular platforms to meet workload-specific demands in high-performance and localized computing environments.

The Rise of AI and GPU-Optimized Systems

AI and GPU-optimized platforms may account for 35% of shipped units in 2030, driven by their dominance in AI training and inference workloads. These systems, with diverse loads ranging from 6-60 kW per system, will represent a critical shift toward high-power-density platforms.

Evolving Role of Storage

Standalone storage platforms are projected to decline as a distinct category. Storage functionalities are increasingly integrated into multi-functional platforms like AI/GPU-optimized, cloud-optimized and heterogeneous compute systems, reflecting the industry’s shift toward unified, workload-specific solutions. The increasing power requirements will likely shift the demands towards 0.8-6 kW.

Shift from Traditional Networking to HPC Networking

Traditional networking platforms are expected to decline as their functionalities become increasingly integrated into modular and heterogeneous systems. Meanwhile, HPC networking will expand to meet the rising data transfer demands of high-density and AI platforms. This transition reflects the growing need for high-performance interconnects capable of managing the data intensity and bandwidth requirements of next-generation workloads, driving HPC networking power demands to 1.5-4 kW, while traditional networking systems remain at lower power levels (<1 kW).

Specialized Systems for Emerging Technologies

Quantum, RISC-based architectures and other domain-specific systems will emerge as niche but critical segments by 2030. These platforms will exhibit a wide variety of power densities, form factors and designs. For example, edge-deployed RISC systems optimized for IoT could operate as efficiently as <100 W, while high-performance platforms for quantum simulations using GPUs or similar processors may demand over 10 kW per node.

Growth in Edge Computing

Edge systems will stand apart due to their localized deployments, latency-sensitive workloads and operation in constrained environments. They are designed for real-time processing in industries like retail, healthcare and autonomous vehicles. Power densities will range from 0.5 kW for IoT gateways to 5 kW for high-performance AI inference platforms.

Emergence of Heterogeneous Compute

Heterogeneous systems integrating CPUs, GPUs and accelerators are currently embedded within other categories, such as AI-optimized and HPC systems. These are expected to comprise an increasing part of the ecosystem, addressing the need for flexible, high-performance computing across a range of workloads with power densities ranging from 1.5 – 6 kW.

In the upcoming years, there will no longer be a one-size-fits-all approach to cooling. The increasing diversity of workloads and hardware platforms necessitates equally specialized cooling solutions tailored to meet the unique thermal demands of each application. This evolution calls for the development of application-optimized cooling systems to ensure performance, efficiency, operations and sustainability in next-generation data centers.

Traditional air cooling

Will retain its relevance, especially for systems with lower power densities, high interface requirements, or frequent maintenance needs. Networking equipment and legacy general-purpose servers, for instance, remain well-suited to air cooling due to its cost-effectiveness and operational simplicity. While its penetration in new deployments may decline, it will continue to serve as a cornerstone for specific segments of the ecosystem. Although its share will decrease in terms of new equipment which relies on air cooling with 45% in 2030, eventually it will stabilize around 30-35%.

Cold plate cooling

Will play a pivotal role in addressing the thermal demands of high-power xPU components like CPUs and GPUs. With precision-targeted cooling capabilities and the ability to efficiently interface with neat chip surfaces, cold plates are indispensable for AI and HPC workloads, where chip power densities surpass the limits of air cooling. Frequently paired with door heat exchangers to manage residual heat, cold plates are expected to dominate high-performance compute environments with approximately 45% of new IT equipment in 2030 being equipped with various types of this technology. Additionally, they will complement immersion cooling solutions to handle the most extreme chip power requirements effectively.

Immersion cooling

Is emerging as the ideal solution for environments characterized by distributed heat loads and high-density deployments. Its ability to manage system-wide thermal challenges makes it particularly well-suited for compact platforms, such as edge computing and for ultra-high-density AI and ML systems where the overall thermal load exceeds the air cooling capabilities. Innovations like immersion-precision cooling, which incorporates directed flow technology, further enhance its capability to handle escalating thermal loads efficiently. We can expect approximately 30% of all new IT equipment in 2030 to rely on some form of immersion cooling technology.

Hybrid cooling solutions

Combining the strengths of various cooling solutions (e.g., cold plates within immersion systems), represent a next-generation approach to managing complex thermal demands while improving performance. These systems merge the pinpoint accuracy of cold plates Door Heat Exchangers and with the broad thermal management coverage of immersion cooling, enabling the highest-density workloads, especially for AI and other high density workloads. Hybrid systems are expected to become indispensable for handling the most extreme power densities and diverse thermal profiles.

By examining the historical and current thermal power requirements of various components, we assess how platform density, measured as total power consumption per server excluding xPUs, has developed over time.

This accumulated data shows that platform density, particularly the cumulative thermal load from non-xPU components, has grown considerably as each generation of components brings higher power and thermal requirements. Plotting these cumulative power demands over time provides a clear visualization of how platform density has evolved to date, marking an ongoing shift toward higher total system loads. Unlike traditional cooling methods, which are limited in scope to specific high-power components, immersion cooling is uniquely equipped to manage this broad, system-wide thermal demand that outpaces traditional cooling’s capabilities.

Power Delivery Losses

Voltage Regulation Modules (VRMs) and Power Delivery Networks (PDNs) can add 66W for a 500W CPU and up to 157W for a 1200W GPU, representing non-negligible heat sources in high-density platforms.

Network Components

With SmartNICs, FPGAs, and high-bandwidth Ethernet cards drawing between 100W and 250W per port, the network layer alone imposes significant thermal demands.

Memory Modules

High-bandwidth memory (HBM) modules and DDR memory have also become major heat contributors, consuming 100W per HBM module and up to 320W across multiple DDR modules.

Storage and Power Supply

Storage accelerators and NVMe SSDs add heat with upto 30W+ per drive, and even the most efficient PSUs contribute to the system’s thermal load with around 53W per kW of total system power.

This particular analysis is outdated (early 2025). Up-to-date forecasts and analysis are exclusively available for Promersion Clients.

By anchoring forecasts to the physical limitations of hardware and the differentiated demands of specific applications, the Promersion model avoids the main pitfalls of traditional market analyses. It captures the inflection point at which liquid cooling transitions from optional to mandatory, ensuring that adoption is not artificially capped by historical patterns.

Cold plate cooling

• Cold plate major take-off in 2024/2025 due to thermal requirements
• 2025 on-track, supply chain rapidly scaling
• Longer term market growth accompanied by reduced cost, lowering revenue/SKU, dampening revenue growth with growing unit sales
• Cold plate becoming the norm, increasingly integrated into thermal packaging, reducing the growth of independent cold plate technologies

Immersion cooling

• Reduced growth over Q2-3 2025 (20% reduced to 3% growth)
  -Impact from cold plate adoption event, new systems pre-equipped with cold plate
• Growth impact most likely temporary in nature, expected ramp-up again in 2026
  -2028 inflection point analysis more broadly recognized
  -Greatly increased OEM involvement in immersion technology
  -More strategic hyperscale pilots/studies

5-Year forecast

• Market evolving to highly diversified IT equipment landscape
• Highlights infliction point in 2024 (Cold plate) and 2028 (Immersion)
• Cold plate to reach plateau in 2030 as addressable market saturates
• Forecasted CAGR (2024-2030) of 47% (cold plate) and 71% (immersion)

These visuals originate from a case study exploring strategic positioning in a rapidly evolving and disruptive ecosystem for a global manufacturer of copper products, Wieland. While grounded in the context of liquid cooling and data center infrastructure, the underlying questions are broadly applicable across industrial and technology-driven markets.

The graphs highlight a core strategic insight: supporting a large ecosystem often offers significantly greater long-term potential and lower structural risk for Wieland than attempting to become a full solution provider. For many manufacturers, value is more sustainably created by enabling scale across the ecosystem, rather than competing on agility, integration, and speed, which are typically the strengths of specialist solution vendors.

By mapping ambition, capability alignment, and execution risk, the framework illustrates why downstream expansion can introduce complexity and dilution, while ecosystem support allows organizations to compound their strengths, remain capital-efficient, and participate in market growth without assuming disproportionate operational risk.