The hidden power systems behind AI [partner content]

Episode Summary

Executive Summary: Scholz Technologies executives Kishin Panadurai and Kamen Berg explain how AI data centers are reshaping electrical infrastructure. They argue batteries are moving from small UPS backup rooms to large external systems that can buffer volatile AI loads, support the grid, and enable new DC-centered architectures. The discussion highlights rapid market change, multiple battery chemistries, and a push toward 800V and eventually 1500V DC distribution.

Main Topics: Batteries are becoming core infrastructure for AI data centers (Priority: 5/5): The guests describe a shift from small UPS batteries used only to bridge generator startup to larger external battery systems that handle load volatility, protect compute footprint, and support grid stability. Wild West market dynamics in data center power (Priority: 5/5): Customers are changing architectures quickly, standards are unsettled, and suppliers must adapt in near real time to evolving AI campus designs and load requirements. From AC-heavy systems to higher-voltage DC architectures (Priority: 5/5): The conversation details how current data centers involve multiple AC/DC conversions, and how moving to 800V DC or 1500V DC could reduce losses and improve efficiency at gigawatt scale. Battery chemistries and footprint tradeoffs (Priority: 4/5): Iron-air, sodium, and vanadium flow batteries are discussed as alternatives to lithium-ion, with emphasis on footprint, installation method, duration, and scalability rather than chemistry alone. Coordination across hyperscalers, contractors, and vendors (Priority: 4/5): Large data center builds involve many specialized subcontractors, making communication among energy, construction, and design teams essential to avoid siloed decisions. Scholz product strategy for DC distribution (Priority: 4/5): Scholz is positioning its cable bus, Power Hub, and related distribution products to serve both UPS and rack-level DC delivery in 800V/1500V ecosystems. Data centers as grid resources, not just loads (Priority: 4/5): The guests argue that mega-scale data center sites should potentially serve both behind-the-meter and front-of-meter roles, helping the grid instead of only consuming from it.

Key Arguments: Legacy UPS rooms are too small and too limited to handle the power volatility of modern AI data centers. External battery systems with higher power density can preserve valuable floor space for compute while providing stronger ride-through capability. Battery storage is now a necessity for AI campuses, not an optional add-on, because grids and diesel generators cannot absorb the load swings alone. The same site may house both data center backup energy and grid-serving batteries; separating those functions is increasingly inefficient. Data center projects require many specialized contractors, so decision-makers must stay aligned to prevent siloed technical choices. Non-lithium chemistries are gaining interest, but the key question is whether they can scale repeatably and economically. DC-native batteries and DC distribution are better aligned with emerging 800V and 1500V architectures because they can reduce conversion losses. Even a small efficiency gain matters enormously at gigawatt scale, making fewer power conversions strategically important. The industry lacks standardization; power block sizes, architectures, and installation approaches vary widely by project. Scholz believes its existing cable bus and DC distribution products can bridge current and future architectures, including higher-voltage systems.

Data Points: Scholz experience in electrical infrastructure: 30 years - Company history cited in the episode sponsor segment. Utility-scale deployment of cable bus material: almost 100 gigawatts - Kamen Berg said the company has deployed nearly 100 GW worth of material into the market. Typical AI data center size: gigawatt-scale - Kishin Panadurai described modern AI data centers as large gigawatt-scale facilities. Example site battery allocation: 250 megawatts - Kamen used this as an example of batteries attached to a 1 GW data center. Example UPS battery scale on a data center site: two full gigawatts - Kamen said a data center might already have 2 GW worth of battery in a UPS system. Power conversion efficiency: around 90% to 98.7% - Kishin described inverter/converter efficiencies and the losses from multiple conversions. Potential efficiency loss at gigawatt scale: 2.5% to 3% - Kishin said even small losses become very significant at gigawatt scale. Common solar DC block size: about 5 megawatts - Kamen compared data center DC blocks to more standardized solar inverter blocks. Non-traditional battery DC block range: 500 kilowatts to 3 megawatts - Kamen said block sizes vary widely based on original design. Representative power outage example: 300,000 people - Kamen cited a winter ice storm in Nashville affecting this many people. Rising rack power example: 10 racks / 100 kilowatts vs. 1 megawatt rack - Kamen described the shift toward much denser rack-level power delivery.

Pivotal Quotes: "I build product that allows the movement of energy, very simply put." — Kishin Panadurai: He is explaining his role in product management for battery storage and power infrastructure. "It's the Wild West because there's a lot of solutions to a problem, and we're at the infancy stage right now of like which one is the right one." — Kamen Berg: He characterizes the current data center power market as fast-moving and unsettled. "If we can cut down the conversion loss or the inversion loss, you can drastically improve your power quality over time." — Kishin Panadurai: He is discussing the benefits of moving toward DC architectures and fewer power conversions.

Implications: AI data centers are becoming massive power systems as much as computing facilities. Expect more external batteries, higher-voltage DC, tighter vendor coordination, and growing pressure to make campuses serve the grid as well as the load.