Searcharxiv⌕ Search

arXiv · 2611.21982

Emerging Vertical Power Delivery in AI Data Centers: Challenges and Solutions

Abstract

The rapid growth of artificial intelligence (AI) workloads is increasing power demand in AI data centers, and processor current is projected to reach 2000 A and beyond. This trend is exposing the limitations of lateral power delivery (LPD) to processors and has motivated growing interest in vertical power delivery (VPD). This paper reviews the power delivery network (PDN) of AI data centers and the upcoming transition from LPD to VPD. First, the 800 VDC power delivery architecture, the impedance behavior of the PDN, and the limitations of LPD are discussed. Then, VPD implementation approaches, reported converter designs, and packaging technologies are examined, together with integrated voltage regulators (IVRs) and embedded passive components. In addition, power integrity simulations are presented to compare regulator and plane positions, from which a set of PDN design guidelines is derived. The results show that moving a voltage regulator module (VRM) from 50 mm away to directly below the die reduces the loop inductance by nearly half. In a complete PDN, this relocation lowers the low-frequency anti-resonant peak, whereas the mid- and high-frequency impedance depends on the decoupling stages and the position of the power and ground planes. This paper also shows that achieving practical VPD remains closely tied to challenges such as regulator height, passive-component embedding, and thermal management.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Charles Isaac Zahner, Amirhossein Habibi, Mehdi Ferdowsi. 2026-11-16. Emerging Vertical Power Delivery in AI Data Centers: Challenges and Solutions. https://searcharxiv.com/abs/2611.21982

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Mean-Field Learning for Storage Aggregation

Distributed energy storage devices can be aggregated to provide operational flexibility for power systems. Their efficient integration into system-level optimization calls for an accurate and computationally tractable surrogate that jointly represents aggregate flexibility and operating costs. However, constructing such a surrogate is challenging due to device heterogeneity, nonconvex operating constraints, and high-dimensional power trajectories. To address these challenges, this paper develops a mean-field learning framework for storage aggregation. Using random-set expectations, we show that the joint representation of aggregate flexibility and operating costs converges almost surely to a deterministic convex mean-field limit. This limit provides a learning benchmark for storage aggregation, against which we bound the approximation error of any convex surrogate. These theoretical results motivate a convex virtual-battery surrogate with physically interpretable parameters. We learn these parameters through a bilevel optimization problem that matches the optimal values of the surrogate and the storage fleet when both respond to a set of price scenarios. We derive closed-form gradients to solve this problem efficiently. Case studies illustrate the theoretical results and demonstrate the framework's approximation accuracy, computational efficiency, and robustness to device heterogeneity.

eess.SY↗

HyperKKL: Enabling Non-Autonomous State Estimation through Dynamic Weight Conditioning

This paper proposes HyperKKL, a novel learning approach for designing Kazantzis-Kravaris/Luenberger (KKL) observers for non-autonomous nonlinear systems. While KKL observers offer a rigorous theoretical framework by immersing nonlinear dynamics into a stable linear latent space, its practical realization relies on solving Partial Differential Equations (PDE) that are analytically intractable. Current existing learning-based approximations of the KKL observer are mostly designed for autonomous systems, failing to generalize to driven dynamics without expensive retraining or online gradient updates. HyperKKL addresses this by employing a hypernetwork architecture that encodes the exogenous input signal to instantaneously generate the parameters of the KKL observer, effectively learning a family of immersion maps parameterized by the external drive. We rigorously evaluate this approach against a curriculum learning strategy that attempts to generalize from autonomous regimes via training heuristics alone. The novel approach is illustrated on four numerical simulations in benchmark examples including the Duffing, Van der Pol, Lorenz, and Rössler systems.

eess.SY↗

Dynamic Flexibility Requests in Local Flexibility Markets: Quantifying the DSO Willingness to Pay

Local Flexibility Markets (LFMs) require Distribution System Operators (DSOs) to determine both the quantity of flexibility to procure and the corresponding willingness to pay during market clearing. Existing approaches typically rely on centralized AC-OPF clearing algorithms with limited practical scalability or strictly localized, static flexibility requests driven primarily by congestion management, while the economic value of flexibility is largely neglected. This paper proposes a dynamic flexibility-request methodology in which the DSO's willingness to pay is embedded directly into the market-clearing objective by monetizing transformer and cable aging, network losses, and voltage congestion. To enable computationally efficient clearing, exact convex piecewise-linear epigraph reformulations of the IEEE C57.91 transformer aging model and an Arrhenius-based cable aging model are developed and formally proven to preserve exactness. The proposed framework is validated on a modified CIGRE MV benchmark and compared with a recent state-of-the-art flexibility-request methodology. The results demonstrate significantly higher market liquidity, more efficient flexibility procurement, and improved network operations while preserving the market's transparency and non-discrimination principles.

eess.SY↗