The paper “AI Hardware Development and Its Consequences for Passive Electronic Components” was presented by Tomas Zednicek, EPCI European Passive Components Institute, Lanskroun, Czech Republic at the 5th PCNS Passive Components Networking Symposium 9-12th September 2025, Seville, Spain as paper No. AI 3.
Introduction
Artificial intelligence (AI) is reshaping the technology landscape with unprecedented computational demands across data centers and edge deployments.
While much attention is given to AI processors and accelerators, the supporting electronics—especially passive components such as multilayer ceramic capacitors (MLCCs), inductors, and resistors—have become critical to sustaining performance, power efficiency, and thermal stability.
AI hardware evolution is driving a shift in power architectures from traditional 12V to 48V and eventually 800V systems, necessitating advanced cooling solutions, high-efficiency energy storage, and precision components to manage extreme transient conditions and high-frequency operations.
Key Points
- AI data centers consume up to eight times more energy than conventional facilities, demanding innovative power management strategies.
- Transition to 48V and forthcoming 800V power supply topologies significantly reduces conversion and distribution losses.
- Passive components must evolve to support high current transients, low parasitics, and thermal stability in extreme environments.
- Memory architecture move from DDR4 to DDR5 raise voltage and decoupling components selection.
- Advanced cooling methods (direct liquid and immersion) and predictive thermal management are essential for high-density AI chips exceeding 1kW per device.
- Emerging technologies include high-cap MLCCs, silicon capacitors, composite and single-turn inductors, and precision thin-film resistors.
- Supercapacitors play a vital role in grid-level energy stabilization for dynamic AI workloads.
Extended Summary
The evolution of AI hardware is characterized by a rapid increase in energy consumption and processing intensity. Specialized compute engines like Google TPU, AWS Trainium, and Nvidia’s Blackwell GPUs are moving away from general-purpose CPUs and GPUs, emphasizing high throughput and low-latency architectures. These systems draw extraordinary power, with individual GPUs exceeding 1kW and full server racks reaching multi-kilowatt thermal design power requirements. This surge in demand has placed significant stress on data center infrastructures and the passive components that underpin power delivery and signal integrity.
Power management in AI systems has undergone a fundamental transformation. The shift from 12V to 48V rack-level distribution and the planned move toward 800V topologies reduce distribution losses, improve efficiency, and prepare data centers for megawatt-scale AI workloads. Adaptive Voltage Scaling (AVS) and Dynamic Voltage and Frequency Scaling (DVFS) complement these architectures by optimizing energy usage in real time. Supercapacitor arrays and energy storage systems stabilize transient loads and improve grid reliability, particularly when integrating renewable energy sources.
Thermal management has become a crucial element of AI infrastructure. Traditional air cooling is insufficient for chips exceeding 1kW, prompting adoption of direct liquid cooling and immersion techniques. Predictive thermal strategies, supported by machine learning, enable proactive adjustments to maintain component longevity and efficiency. These thermal considerations directly impact the selection and performance of passive components near high-heat sources.
Passive component engineering is advancing to meet these challenges. MLCC capacitors are now designed with ultra-low ESR and ESL for near-die decoupling in AI servers. High-voltage and automotive-grade MLCCs, as well as silicon capacitors with sub-pH parasitics, support stable high-current operations. Controlled ESR polymer tantalum capacitors and aluminum electrolytic capacitors enhance stability in feedback-sensitive circuits, while supercapacitors can manage large energy bursts and stabilize the data center power grid.
Inductor technology is evolving with single-turn and composite core designs that handle extreme currents, minimize losses, and maintain thermal stability. These inductors enable compact, efficient power delivery for AI accelerators and high-frequency switching regulators. Precision resistors with tight tolerances and low noise, including thin-film and metal foil types, ensure signal integrity in analog and high-speed digital systems. Innovations in programmable resistors and memristor-based devices hint at future pathways for in-memory and neuromorphic computing to reduce AI power consumption.
Conclusion
AI hardware development has triggered a paradigm shift in data center power and thermal design, demanding a new generation of passive components. The combination of higher voltage topologies, advanced cooling, and dynamic power management strategies ensures that AI systems can scale sustainably.
Continuous innovation in capacitors, inductors, and resistors will remain essential to meeting the extreme demands of AI workloads. A multidisciplinary approach—linking power architecture optimization, passive component engineering, and predictive thermal management—will be the foundation of future high-performance, energy-efficient AI infrastructure.