The transition from 54-V to 800-V energy in AI information facilities

October 29, 2025

3

The transition from 54-V to 800-V energy in AI information facilities

Whereas compute gadgets equivalent to CPUs, GPUs, and XPUs are stealing the limelight within the synthetic intelligence (AI) period, there may be an growing realization that powering AI at scale calls for new energy techniques and architectures. In different phrases, information heart operators are investing closely in high-performance computing for AI, however there is no such thing as a AI with out energy.

The exponential progress of AI is quickly outstripping the capability of the present 54-V information heart energy infrastructure, driving a change towards high-density, dependable, and protected 800-V powered information facilities. Right here, at this know-how crossroads, the brand new energy supply structure requires new energy conversion options and security mechanisms to forestall potential hazards and dear server downtimes.

Determine 1 AI information heart energy was a outstanding theme at Infineon OktoberTech Silicon Valley 2025. Supply: Infineon

At Infineon’s OktoberTech Silicon Valley occasion held on 16 October 2025 in Mountain View, California, this tectonic shift in information heart energy infrastructure was a significant spotlight. The corporate demonstrated 800-V AI information heart energy architectures constructed round silicon, silicon carbide (SiC), and gallium nitride (GaN) applied sciences.

Infineon has additionally joined fingers with Nvidia to maximise the worth of each watt in AI server racks by modular and scalar energy architectures. The 2 firms will work collectively on information heart energy facets, equivalent to hot-swap controller performance, which allows future server boards to function in 800-V energy architectures. It can facilitate the alternate of server boards on an 800 VDC bus whereas your complete rack continues working by managed pre-charging and discharging of the boards.

At Infineon OktoberTech Silicon Valley, Peter Wawer, division president of inexperienced industrial energy at Infineon Applied sciences, spoke with EDN to clarify the transition to AI information facilities to 800-VDC architectures. He additionally walked by the demo to point out how 800-V energy is delivered to AI server racks.

The arrival of solid-state circuit breakers

“We’re seeing a change to an 800-VDC structure in AI information facilities, which is a significant step ahead to establishing highly effective AI gigafactories of the long run,” Wawer stated. “The facility consumption of an AI server rack is estimated to extend from round 120 kilowatts to 500 kilowatts, and to 1 megawatt by the top of the last decade.”

Inevitably, it requires increased effectivity and lowered losses as computing energy continues to scale at an unprecedented charge. “This evolution brings new challenges,” Wawer acknowledged. “Whenever you need to alternate server boards on an 800-V bus whereas your complete rack continues working, you’re coping with substantial energy ranges.”

For example, engineers want managed pre-charging and discharging to keep away from harmful inrush currents and guarantee protected upkeep with out downtime. Whereas conventional protecting gadgets like fuses and mechanical breakers have served reliably for many years, they weren’t designed for the ultra-fast fault response required in as we speak’s high-voltage, high-speed environments, the place microseconds matter.

That’s the place the subsequent era of solid-state circuit breakers (SSCBs) is available in. The brand new information heart architectural shift is resulting in the emergence of SSCBs, which can modernize AI information facilities whereas changing electromagnetic transformers. SSCBs reply to faults in microseconds with very excessive precision, which makes energy distribution in AI information facilities safer, sooner, and extra environment friendly.

Determine 2 SSCBs will change electromagnetic transformers that presently join the grid to energy infrastructure in information facilities. Supply: Infineon

“To allow these next-generation SSCBs, Infineon launched the CoolSiC JFET household earlier this yr,” Wawer advised EDN. “These JFETs supply the power to mix ultra-low on-resistance—1.5 mΩ at 750 V and a couple of.3 mΩ at 1200 V—to make sure strong efficiency even below powerful circumstances.”

Reliability is one other key benefit, he added. “These JFETs are designed to deal with sudden voltage spikes and present surges, responding rapidly to faults and serving to stop gear harm or downtime.” Their packaging—aided by top-side cooling and Infineon’s .XT interconnect know-how—helps AI information heart energy techniques keep cool and dependable even in probably the most demanding environments.

These JFETs additionally scale back the necessity for exterior clamping circuits, simplifying system design and enabling extra compact and cost-effective options. Apart from AI information facilities, this SSCB know-how can assist defend electrical autos (EVs), industrial automation and good grids, making energy distribution safer, extra environment friendly, and prepared for the long run.

Stable-state transformers, hot-swap controllers, and energy modules

At OktoberTech Silicon Valley, Infineon additionally demonstrated an influence system constructed round high-voltage CoolSiC parts for high-voltage DC energy distribution to IT racks powered by a solid-state transformer (SST). “The SSTs can be essential in gigawatt-scale AI datacenters,” Wawer stated.

An SST is a power-electronics stack for connecting the grid to information heart energy distribution. It replaces the standard techniques based mostly on a low-frequency transformer manufactured from copper and metal and an AC-DC converter, enabling a dramatic discount in measurement and weight, end-to-end effectivity, and lowered CO₂ footprint.

Subsequent, Infineon unveiled a reference board for hot-swap controllers for 400-V and 800-V energy architectures in AI information facilities. The recent-swap controller performance is important to offering the very best ranges of safety, maximizing server uptime, and making certain optimum efficiency. The REF_XDP701_4800 hot-swap controller reference design is optimized for future 400-V/800-V rack architectures.

Determine 3 Sizzling-swapping controller designs demonstrated at OktoberTech in Silicon Valley are optimized for 400-V/800-V information heart rack architectures. Supply: Infineon

Then there have been trans-inductance voltage regulator (TLVR) modules particularly designed for high-performance AI information facilities. Infineon’s TDM22545T modules mix OptiMOS know-how energy levels with TLVR inductors to bolster energy density, enhance electrical and thermal effectivity, and improve sign high quality with lowered transients.

The proprietary inductor design delivers ultra-fast transient response to dynamic load modifications from AI workloads with out compromising electrical or thermal effectivity. Furthermore, the inductance structure minimizes the variety of output capacitors, decreasing the general measurement of the voltage regulator (VR) and decreasing bill-of-materials (BOM) prices.

Determine 4 The TLVR modules ship benchmark energy density and transient response essential in AI information facilities. Supply: Infineon

Transition to new energy architectures

Jim McGregor, principal analyst at Tirias Analysis, acknowledges that it’s turning into more and more difficult to energy AI information facilities from the grid to the chip degree. “It’s essential that energy design engineers repeatedly enhance effectivity, energy density, and sign integrity of energy conversion from the grid to the core.”

Particularly when an AI server prices 30 instances as a lot as a standard server. Moreover, there may be an growing have to simplify system design, enabling extra compact, cost-effective options for powering AI information facilities.

The approaching shift from the present 54-V information heart energy infrastructure to a centralized 800-V structure is a part of this design journey within the quickly evolving world of AI information facilities. That inevitably calls for brand new constructing blocks—hot-swap controllers, SSCBs, and SSTs—to efficiently migrate to new energy architectures.

These power-electronics constructing blocks at the moment are obtainable, which implies the transition to 400-V/800-V AI information facilities isn’t far off.

Associated Content material

The publish The transition from 54-V to 800-V energy in AI information facilities appeared first on EDN.