Bottlenecks in AI Advancement: Power Infrastructure and Transformer Contacts

The rapid advancement of AI technology is shifting the core bottleneck from GPU- and high-performance server-centric infrastructure toward power supply systems and the efficiency of power conversion and distribution, which increasingly determine performance and cost. In large-scale environments where AI training and inference operate continuously, not only total power consumption but also cumulative losses (heat) generated across every stage of power transmission and conversion significantly expand facility scale and operating expenses. As a result, data center expansion is no longer a matter of simply adding servers, but a challenge of redesigning the entire power network, from substations and distribution to power conversion and down to rack- and board-level delivery.

This challenge can be summarized in three key points. First is the absolute amount of power. AI workloads, characterized by high computational density and utilization rates, create near-constant high-load power profiles, while site-level power contracts, intake capacity, and substation infrastructure limit expansion speed. Second is heat and cooling. Power losses directly translate into heat, and as losses increase, cooling infrastructure—such as HVAC, liquid immersion, or direct liquid cooling—must scale accordingly. Third is efficiency. Power passes through multiple stages of conversion (AC/DC, DC/DC) and distribution (PDUs, busbars, connectors, relays/contactors). Even small losses at each stage erode overall efficiency, and in high-current sections, increased contact resistance immediately leads to heat generation and reduced reliability. Ultimately, long-term competitiveness depends not only on server performance, but on how efficiently and reliably power can be delivered.

From this perspective, electrical contact materials represent a key lever for reducing “invisible losses.” In high-current, high-frequency switching environments, contacts must maintain consistently low contact resistance while resisting degradation mechanisms such as oxidation, arcing, wear, welding, and temperature rise. Contact performance is not determined by conductivity alone. Under real operating conditions, thin surface oxide films, variations in contact pressure, vibration, and surface roughness changes caused by repeated switching all affect resistance. These fluctuations generate heat, which in turn accelerates oxidation and softening, creating a vicious cycle. Therefore, a balanced combination of high conductivity, arc resistance, oxidation resistance, and mechanical stability is essential.

The direction of high-efficiency contact materials is clear. First, it is effective to build on highly conductive bases such as silver or copper, while precisely engineering functional phases—through oxide dispersion, alloying, or composites—that resist degradation. Second, in long-term operation and repeated switching, stability of contact resistance over the entire service life is becoming more important than its initial value. Third, as current levels increase, system-level optimization becomes essential, encompassing not only materials but also surface treatments (plating/coatings), joining methods (brazing), and structural design (contact geometry, spring and pressure mechanisms). Electrical contact materials are therefore being re-evaluated as material-based solutions that reduce micro-losses and enhance reliability across data center power networks.

1. Stages of AI Development and Infrastructure Demand

(1) Initial build-out phase (present–2027)
(2) Expansion and optimization phase (2028–2032)
(3) Sustainability enhancement phase (2033 and beyond)

Initial build-out phase: As AI model scales expand—such as large language models and multimodal AI—data center computing capacity grows rapidly. At this stage, infrastructure demand focuses on servers and networking equipment, but power consumption is already emerging as a critical issue. As of 2024, global data centers account for approximately 4% of total electricity consumption, a figure expected to more than double by 2030. This corresponds to an annual demand of roughly 900–1,000 TWh, comparable to the electricity consumption of a mid-sized country.

Expansion and optimization phase: As AI applications spread across industries, data center density increases. Here, power supply stability and efficiency become decisive factors. Demand for power conversion equipment such as transformers surges, and energy consumption for AI training and inference can account for 30–50% of total operating costs.

Sustainability enhancement phase: Integration of renewable energy and efficiency technologies becomes essential. However, aging power grids and delays in power supply expansion pose significant risks that could slow AI growth. In this phase, improving the efficiency of electrical contacts within transformers plays an important role in minimizing energy losses.

2. Power Challenges After AI Infrastructure Deployment and Rising Transformer Demand

Once initial AI infrastructure is deployed, power supply becomes the greatest challenge. Data center power demand is projected to increase by 50–165% by 2030, and in the United States alone, capacity is expected to reach 78–123 GW by 2035. This exceeds existing grid capacity and drives a sharp increase in demand for transformers, particularly large power transformers.

Transformers are critical equipment that convert high-voltage electricity into lower voltages suitable for data centers. In AI data centers, transformers are responsible for maintaining voltage stability and minimizing energy losses. Due to rapid data center expansion, transformer demand increased by 274% between 2019 and 2025. Global supply chain constraints have extended transformer lead times to two to three years, becoming a major cause of AI project delays. High-capacity transformers, such as generation step-up transformers (GSUs), are heavily import-dependent (approximately 80% in the United States), exposing projects to additional risks from trade policies and raw material price volatility.

3. The Role of Electrical Contacts in Transformers and Bottlenecks in AI Advancement

Within transformers, electrical contacts are primarily used in on-load tap changers (OLTCs). OLTCs allow transformers to dynamically adjust voltage under load, with contacts controlling current flow and maintaining stability during arcing events. The main functions of these contacts include:

Current switching: Ensuring stable current transfer in high-voltage environments (tens of kV) to improve transformer efficiency.
Minimizing energy loss: Low contact resistance reduces heat generation and energy waste. While overall transformer efficiency exceeds 95%, inefficiencies at contact points can account for 10–20% of total losses.
Ensuring durability: Contacts must withstand wear and arc erosion caused by repeated switching, often thousands of operations per year.

4. High-Efficiency Electrical Contact Materials

As AI energy consumption increases, the efficiency of electrical contacts has become a critical issue. Conventional contact materials contribute to energy losses, whereas high-efficiency materials can reduce contact resistance by 20–50%, improving overall transformer efficiency by 1–2%. At data center scale, this translates into annual energy savings of hundreds of gigawatt-hours.

Silver–tungsten (Ag-W): Combining silver’s excellent conductivity with tungsten’s arc resistance, this material is among the most effective in high-voltage environments. Its low contact resistance (0.1–0.5 mΩ) minimizes heat loss, and its long service life (over 100,000 switching cycles) meets the high-load demands of AI systems. Studies show that Ag-W reduces arc erosion rates by approximately 40%, ensuring long-term stability.

Copper–chromium (Cu-Cr): This material leverages copper’s high conductivity and chromium’s corrosion resistance, offering a cost-effective solution. It remains stable under high currents (thousands of amperes) and maintains energy efficiency above 95%. Widely used in OLTCs, it demonstrates strong durability against fluctuating loads in AI data centers.

Gold-plated contacts (Au-plated): These consist of a thin gold layer plated over a base material such as silver or copper, and are particularly effective in low-voltage, low-current applications. Gold’s excellent resistance to corrosion and oxidation enables contact resistance to be maintained at 0.05–0.3 mΩ, improving electrical conductivity. In high-density power systems within AI data centers, gold-plated contacts are suitable for auxiliary contacts or precision switching components, offering protection against fretting wear and enhanced durability. However, due to gold’s softness, wear can occur in high-current environments, so nickel under-plating is often used. Although costly, gold plating reduces long-term maintenance costs and is well suited to high-reliability requirements such as uninterrupted data center operation. Research indicates that gold plating can extend contact life by two to three times and improve energy efficiency by 5–10%.

5. Characteristics of Electrical Contact Materials in Transformers

The future of AI depends on the stability and efficiency of power infrastructure. The expansion of transformer demand and the advancement of contact technologies are key to resolving power bottlenecks. The adoption of high-efficiency electrical contact materials can reduce power losses and improve overall energy consumption by approximately 10–20%. Ultimately, the competitiveness of AI infrastructure will be determined not by computational performance alone, but by how efficiently and reliably power can be delivered.