AI data centers are changing the way large computing facilities use and distribute electricity.
Higher GPU density, larger computing clusters, continuous high-load operation, advanced cooling systems, and rapidly expanding campuses are pushing data center power requirements far beyond those of many traditional facilities. At the same time, new power architectures such as higher-voltage DC distribution, battery energy storage, and solid-state transformers are receiving increasing attention.
One of the most discussed developments is 800VDC power distribution.
By operating at a higher DC voltage, future AI data centers may be able to deliver large amounts of power with lower current, smaller conductor requirements, and fewer conversion stages between the electrical distribution system and high-density computing equipment.
This naturally raises another question:
If AI data centers increasingly move toward 800VDC distribution, do they still need conventional power transformers?
In most large facilities, the answer is yes.
800VDC changes an important part of the downstream power architecture, but it does not remove the need to connect the data center to the utility grid, transform incoming voltage, distribute large amounts of power across the site, maintain electrical isolation, and support redundant electrical infrastructure.
Solid-state transformers may eventually change some conversion stages inside the data center, but conventional power transformers are still likely to remain an important part of the overall electrical system.
As a power transformer manufacturer, we see the development of 800VDC and solid-state conversion less as the disappearance of the conventional transformer and more as a change in where different transformation and conversion technologies are used within the same data center.
Why Is 800VDC Attracting Attention in AI Data Centers?
The amount of electrical power required by AI computing is increasing quickly.
Traditional data centers were largely designed around CPU-based workloads. Modern AI training and inference systems rely heavily on dense GPU clusters and other accelerators, creating significantly higher power demand within individual racks and data halls.
As rack power increases, supplying the same amount of power at relatively low voltage requires increasingly high current.
Higher current can mean:
- larger conductors;
- greater cable losses;
- more heat;
- more demanding busbar and connector requirements;
- additional space for power distribution equipment.
Increasing the distribution voltage is one way to reduce these pressures.
This is why higher-voltage DC architectures, including 800VDC distribution, have become increasingly relevant to next-generation AI data center designs.
Another consideration is the number of conversion stages between the grid and the computing equipment.
A conventional data center power path may include several stages:
Utility Supply → Transformer → Medium-Voltage Distribution → Low-Voltage Distribution → UPS / Rectifier → DC Power Supply → IT Load
Every stage serves a purpose, but every additional conversion stage also introduces equipment, heat, space requirements, controls, and electrical losses.
New AI data center architectures are therefore exploring ways to simplify parts of this chain.
However, simplifying downstream conversion is very different from eliminating transformers throughout the entire facility.
Where Does a Conventional Power Transformer Fit in an AI Data Center?
Before looking at the server rack, it is useful to look at how electricity actually enters a large data center.
A large AI campus may receive electricity from the utility at medium voltage or high voltage. Some projects may also use dedicated substations, multiple utility feeds, on-site generation, renewable energy systems, or battery energy storage.
Whatever downstream architecture is selected, the incoming electrical supply still has to be matched to the voltage levels used across the site.
A simplified arrangement may look like:
Utility Grid → Substation → Power Transformer → Medium-Voltage Distribution → Data Hall Power Conversion → IT Load
The exact voltage levels vary between projects, but the fundamental requirement remains.
Grid Voltage Still Has to Be Transformed
An AI data center cannot simply choose its internal operating voltage independently of the external utility network.
If the utility supplies electricity at a voltage that is not suitable for site distribution, voltage transformation is required.
This is one of the most established roles of the conventional power transformer.
Depending on the scale of the project, transformers may be used between high-voltage transmission or sub-transmission networks and the campus medium-voltage system, or between different medium- and low-voltage distribution levels.
Even if the final distribution toward the computing racks moves to 800VDC, the upstream grid connection does not disappear.
Large Campuses Still Need Bulk Power Distribution
AI campuses can require very large and continuously available electrical capacity.
This is different from converting several hundred kilowatts inside an individual rack or power block.
At campus level, the electrical system may need:
- multiple transformer units;
- separate substations;
- redundant utility feeds;
- independent power blocks;
- reserve transformer capacity;
- staged expansion for future data halls.
Conventional transformers are already widely used for this type of bulk voltage transformation and electrical distribution.
That makes their role particularly difficult to replace simply by changing the voltage architecture closer to the IT equipment.
Electrical Isolation and Protection Remain Part of the System
Voltage transformation is only one part of transformer application.
Electrical isolation, grounding arrangements, protection coordination, fault management, impedance, and the relationship between the transformer and surrounding switchgear all influence how a data center electrical system operates.
Higher-voltage DC distribution can change the downstream architecture, but these upstream engineering requirements still need to be addressed.
Where Could Solid-State Transformers Enter the Power Chain?
A solid-state transformer, or SST, performs voltage transformation differently from a conventional power transformer.
A conventional transformer primarily relies on electromagnetic induction.
An SST combines power electronics with transformer isolation and active control. Depending on the design, this can allow voltage transformation and AC/DC conversion to be integrated into the same system.
This creates an interesting possibility for AI data centers.
Instead of a power path such as:
MV AC → Conventional Transformer → LV AC → Rectifier → DC
some future applications may use:
MV AC → Solid-State Transformer / Integrated Power Conversion → DC
Reducing separate conversion stages can be attractive where space, conversion efficiency, power density, and DC distribution are important.
But this does not necessarily mean that the solid-state transformer replaces every conventional power transformer on the site.
In many cases, it may first appear further downstream, between the medium-voltage distribution system and specific high-density computing power blocks.
That distinction is important for understanding how the two technologies may develop together.
800VDC Does Not Automatically Mean Solid-State Transformers
One point is particularly important when discussing future AI data center architecture:
An 800VDC distribution system does not automatically require a solid-state transformer.
800VDC describes a distribution voltage architecture. SST describes one possible approach to transformation and power conversion.
They are related, but they are not the same technology.
An AI data center could theoretically reach an 800VDC distribution bus using different combinations of:
- conventional transformers;
- rectifiers;
- centralized power conversion systems;
- modular conversion equipment;
- solid-state transformers.
Which architecture is selected depends on the project.
This means that even if 800VDC becomes increasingly common, the real question is not simply whether a facility uses SST.
The more practical question is:
Where should voltage transformation take place, where should AC-to-DC conversion take place, and which equipment arrangement provides the required reliability, efficiency, maintainability, and scalability?
For some sections of the electrical system, integration may provide a clear advantage.
For others, a conventional power transformer followed by dedicated conversion equipment may remain a practical solution.
Why Conventional Power Transformers Are Unlikely to Disappear
Solid-state transformers offer capabilities that conventional transformers do not provide on their own.
Active voltage regulation, controlled power flow, integrated AC/DC conversion, and potential bidirectional operation can all be useful in advanced power systems.
But data center infrastructure is not selected only according to how many functions can be integrated into one piece of equipment.
Large Power Levels Favor Proven Equipment
AI campuses may operate at tens or even hundreds of megawatts, with individual projects divided into multiple large electrical blocks.
At this scale, power equipment must support continuous operation, fault management, redundancy, maintenance, and long service life.
Conventional power transformers have decades of operating experience across utility networks, industrial plants, renewable energy facilities, transportation infrastructure, and data centers.
Their manufacturing methods, insulation systems, cooling arrangements, protection practices, testing procedures, and maintenance requirements are already well established.
SST technology continues to develop, but replacing an established high-capacity transformer installation is a very different challenge from introducing power electronics into a smaller downstream conversion stage.
More Active Components Also Mean Different Maintenance Requirements
A solid-state transformer adds semiconductor switching devices, control electronics, sensors, cooling systems, and high-frequency magnetic components.
These allow much greater control over the electrical output.
At the same time, they create a different maintenance environment.
A conventional oil-immersed or dry-type power transformer also requires inspection and maintenance, but transformer service practices are already widely understood across global utility and industrial markets.
This becomes especially relevant for international data center projects where long-term local maintenance capability matters almost as much as initial equipment performance.
Reliability Is Still the Main Requirement
AI computing may be advancing quickly, but the electrical infrastructure behind it still has to operate continuously.
Data centers are therefore normally designed around redundancy, fault isolation, reserve capacity, and planned maintenance.
A new power architecture must demonstrate that it can support these requirements at the scale required by the project.
For this reason, adoption of new transformer technologies is more likely to happen gradually than through immediate replacement of the entire existing power chain.
Conventional Transformers and SSTs May Work Together
The future AI data center is unlikely to be defined by one universal electrical architecture.
Different facilities may use very different combinations of AC distribution, DC distribution, batteries, UPS systems, transformers, and power electronic conversion.
A conventional architecture could continue to use:
Utility Grid → Conventional Power Transformer → MV/LV Distribution → UPS / Rectifier → IT Load
A newer hybrid architecture could use:
Utility Grid → Conventional Power Transformer → MV Distribution → Integrated MV-to-DC Conversion → 800VDC Distribution → AI Racks
Another project may introduce SST technology within selected power blocks.
The important point is that the presence of a solid-state transformer downstream does not necessarily remove the conventional transformer upstream.
The two technologies can perform different roles.
| Area | Conventional Power Transformer | Solid-State Transformer |
|---|---|---|
| Main operating principle | Electromagnetic transformation | Power electronics with transformer isolation |
| Voltage transformation | Yes | Yes |
| Integrated AC/DC conversion | Normally separate equipment | Can be integrated |
| Active power control | Limited without additional equipment | Built into the conversion system |
| Technology maturity | Highly established | Developing |
| Maintenance ecosystem | Widely established | More specialized |
| Likely AI data center role | Grid connection and bulk voltage transformation | Selected conversion and DC power blocks |
For large campuses, this coexistence model may be particularly practical.
The data center can continue using conventional power transformers for the high-capacity upstream network while introducing new conversion technologies where they provide a clear benefit closer to the computing load.
What Changes When Selecting a Power Transformer for an AI Data Center?
Although conventional transformers are likely to remain important, AI data centers can place different demands on the transformer compared with many traditional commercial loads.
The transformer therefore should not be selected only according to total kVA or MVA.
Transformer Capacity and Future Expansion
AI infrastructure can grow quickly.
A campus may begin with several data halls and later add additional GPU clusters, cooling systems, battery storage, or entirely new buildings.
Transformer planning should therefore consider both the initial load and the expected expansion strategy.
In some projects, using several transformer units across independent power blocks can provide greater flexibility than concentrating all capacity into a single unit.
The appropriate arrangement depends on redundancy requirements, site layout, utility supply, future capacity, and maintenance strategy.
Voltage Ratio
The required transformer voltage ratio depends on both the incoming grid and the internal distribution system.
For example, a data center connected to a higher-voltage utility network may require transformation before power enters the campus medium-voltage distribution system.
Other projects may receive power directly at medium voltage and require transformation closer to individual data halls.
The growing use of higher-voltage DC does not remove this consideration. It simply changes where the next conversion stage may occur.
Impedance and Short-Circuit Conditions
Transformer impedance affects fault current, voltage drop, parallel operation, and protection coordination.
In a large data center with multiple transformers and interconnected electrical sections, these factors should be considered together rather than independently.
Where transformer units are expected to operate in parallel, voltage ratio, vector group, impedance characteristics, and other operating parameters also need to be compatible.
Harmonics and Power-Electronic Loads
AI data centers contain a large amount of power-electronic equipment.
Depending on the facility, this may include:
- UPS systems;
- rectifiers;
- server power supplies;
- variable frequency drives;
- cooling equipment;
- battery converters;
- renewable energy inverters;
- future solid-state conversion systems.
These loads can affect harmonic conditions within the electrical network.
Transformer design therefore needs to reflect the actual load characteristics and system configuration instead of relying only on total rated capacity.
Cooling and Continuous Operation
Data center electrical systems normally operate continuously and often at high utilization.
Transformer cooling performance therefore becomes particularly important.
Oil-immersed power transformers and dry-type transformers use different cooling arrangements, and the appropriate choice depends on installation location, voltage level, capacity, fire-safety requirements, environmental conditions, and the wider substation design.
Ambient Conditions
Large AI campuses are now being planned across a wide range of climates.
High ambient temperature, altitude, humidity, dust, coastal environments, and outdoor installation can all affect transformer specification.
For projects in the Middle East and other hot regions, for example, transformer temperature rise and cooling design should reflect actual site conditions rather than assuming mild operating temperatures.
These considerations remain important regardless of whether the downstream distribution system ultimately uses conventional AC architecture, 800VDC, or a combination of both.
HONLE Power Transformers for Evolving AI Data Center Infrastructure
AI data center power architecture will continue to evolve.
800VDC distribution, battery energy storage, higher-capacity power blocks, advanced cooling, and solid-state conversion may gradually change how electricity moves from the campus electrical system toward high-density computing equipment.
But this transition does not mean that the conventional power transformer is becoming irrelevant.
Large facilities still need reliable grid connection, voltage transformation, electrical isolation, medium- and high-voltage distribution, redundancy, and long-term operation.
At HONLE, we manufacture power transformers for utility, industrial, renewable energy, infrastructure, and other demanding electrical applications. For data center projects, we consider the transformer as part of the complete electrical system rather than simply matching a rated capacity.
Voltage ratio, transformer capacity, impedance, load characteristics, cooling method, insulation requirements, installation environment, accessories, redundancy, and future expansion can all influence the final transformer configuration.
As AI data center architecture develops, some downstream conversion stages may change substantially.
Conventional power transformers, however, are likely to remain an important part of the upstream and campus-level power system for many projects.
The future is therefore less likely to be a simple choice between a conventional power transformer and a solid-state transformer.
It is more likely to involve different technologies being used at different points in the same electrical system, with conventional transformers continuing to handle large-scale voltage transformation while advanced power electronics move closer to high-density AI loads.
FAQ
Do AI data centers still need power transformers with 800VDC distribution?
Yes. 800VDC mainly changes how electricity may be distributed closer to high-density computing equipment. AI data centers still need to connect to the utility grid and transform incoming voltage to suitable levels for the campus electrical system.
Will solid-state transformers replace conventional power transformers?
Not completely in the foreseeable future. SST technology may replace a conventional transformer together with separate conversion equipment in selected applications, but conventional power transformers remain well suited to grid connection, substations, bulk voltage transformation, and large-capacity electrical distribution.
Is a solid-state transformer required for an 800VDC data center?
No. An 800VDC architecture can potentially use different combinations of conventional transformers and power conversion equipment. SST is one possible approach, but 800VDC and solid-state transformer technology are not the same thing.
Where are conventional power transformers used in AI data centers?
They can be used in utility interconnection substations, high-voltage and medium-voltage transformation, campus distribution systems, and individual electrical power blocks. Their exact location depends on the incoming grid voltage and the overall data center power architecture.
How do AI loads affect power transformer selection?
AI data centers can involve high continuous loads, extensive power-electronic equipment, redundancy requirements, rapid capacity expansion, and demanding cooling conditions. Transformer capacity, impedance, harmonics, temperature rise, cooling, voltage ratio, and installation environment should therefore be considered together.
Can conventional power transformers and solid-state transformers be used in the same data center?
Yes. A large AI campus could use conventional power transformers for grid-side and bulk voltage transformation while using solid-state transformers or other advanced power conversion equipment further downstream. This type of mixed architecture may become increasingly practical as data centers adopt higher-voltage DC distribution.