Why AI Data Centers Are Moving Toward 800V Power Systems
AI data centers are becoming more demanding electrical environments. As operators deploy dense computing systems for training and inference, the challenge extends beyond installing faster processors. Facilities must deliver large amounts of power reliably, convert it efficiently, distribute it safely, and remove the heat produced by the equipment.
On October 5, 2026, Microchip Technology and Navitas Semiconductor announced an 800V direct-current power reference design aimed at AI data-center rack applications. The design combines digital power control, gallium nitride power devices, and hardware-based security features. Microchip also announced an expanded data-center portfolio on October 12, covering connectivity, storage, and acceleration-related infrastructure.
These developments reflect a broader engineering problem: modern AI infrastructure requires coordination across chips, networking, power conversion, storage, cooling, and system security. Higher-voltage rack architectures are one possible part of the response, but they introduce design and safety requirements that cannot be ignored.
Why AI Infrastructure Needs More Than Faster Chips
An AI accelerator can process enormous amounts of data, but it cannot operate independently of the rest of the data center. It needs a stable power supply, access to memory and storage, high-speed communication with other components, and cooling capable of removing heat.
When a facility adds more accelerators, electrical and thermal demands can grow alongside computing capacity. Power must travel through multiple conversion and distribution stages before reaching the components that use it. Each stage has electrical losses, physical constraints, monitoring requirements, and reliability considerations.
Networking and storage also become important. Accelerators can spend time waiting for data if the surrounding system cannot supply information quickly enough. Poorly coordinated infrastructure may therefore limit the useful performance of expensive computing hardware.
The result is a systems-engineering challenge. Operators must optimize the complete computing environment rather than focusing only on processor specifications.
What Is an 800V DC Rack Architecture?
Direct current, or DC, flows in one direction. Alternating current, or AC, periodically changes direction. Electricity passes through conversion and distribution stages as it moves from a facility's electrical supply toward the low-voltage power rails used by electronic components.
An 800V DC rack architecture uses a high-voltage direct-current distribution stage within the data-center power system. The objective is to support high power density and reduce certain distribution losses and conductor requirements compared with some lower-voltage arrangements.
For a given power level, increasing voltage can reduce current. Electrical resistive losses in a conductor depend on current squared and resistance. Under comparable conditions, lower current can therefore reduce resistive losses and influence the size and thermal requirements of conductors.
However, this does not mean that an 800V design is automatically more efficient in every installation. Overall performance depends on conversion efficiency, the distribution layout, equipment selection, load characteristics, safety systems, and the complete electrical design.
Higher voltage also increases the importance of insulation, isolation, protection, monitoring, maintenance procedures, and safe handling. The engineering benefits must be assessed alongside these requirements.
What Microchip and Navitas Announced
On October 5, 2026, Microchip and Navitas described an 800V DC-to-6V DC reference design for AI data-center rack power applications. The announced platform combines Microchip's digital power-control and hardware-security technologies with Navitas gallium nitride power devices.
The companies said the design is intended to help developers evaluate power conversion aligned with the Open Compute Project's 800V DC direction. They also described hardware-based security using Microchip's TA100 CryptoAuthentication integrated circuit.
A reference design is an engineering starting point, not proof that every data center has adopted the architecture or that every deployment will deliver identical results. Developers still need to validate thermal behavior, efficiency, compatibility, protection, reliability, and compliance within their own systems.
The announcement is nevertheless useful because it demonstrates how semiconductor suppliers are preparing components and design resources for the evolving power needs of AI infrastructure.
Why Gallium Nitride Matters
Gallium nitride, commonly abbreviated GaN, is a semiconductor material used in power electronics. GaN devices can support switching characteristics that make them attractive for certain power-conversion applications.
Power converters switch electrical energy between voltage levels and operating conditions. Their design influences energy losses, heat production, physical size, and the ability to respond to changing loads.
In a high-density data center, improvements to conversion efficiency can matter because even small losses become significant when equipment operates continuously at high power. Reduced conversion losses can also lower the amount of heat that cooling systems must remove.
But material choice alone does not determine the outcome. Converter topology, switching frequency, magnetic components, control algorithms, thermal design, and component quality all affect performance.
GaN is therefore best understood as one enabling technology in a larger electrical architecture rather than a standalone solution to data-center energy demands.
Digital Power Control and Hardware Security
Digital power control allows electronic systems to monitor and regulate aspects of power conversion. Depending on the design, controllers can support measurement, adjustment, fault response, and coordination with the surrounding system.
Security is another consideration because data-center equipment forms part of a complex supply chain. Hardware-based security components can support device authentication and trusted firmware operations, helping system designers establish checks on the identity and integrity of selected components.
Microchip described its TA100 device as supporting functions that include secure boot-related operations, firmware authentication, and key-management capabilities. These features can help build a foundation for trust, but they do not make an entire data center secure automatically.
Security still depends on how components are configured, how credentials are protected, how updates are managed, how systems are monitored, and how operators respond to vulnerabilities.
The Other Bottlenecks: Networking and Storage
Power is only one part of the problem. AI systems frequently need to move large volumes of data between processors, memory, storage, and other servers. As workloads grow, the surrounding network and data-access architecture can influence performance as much as the accelerators themselves.
Microchip's October 12 announcement described an expanded portfolio involving PCI Express connectivity, NVMe storage, and acceleration-related technologies. These areas address different parts of the infrastructure stack.
PCI Express provides a high-speed interface used to connect components in computing systems. NVMe is a protocol designed for accessing non-volatile storage, commonly over PCI Express. Their performance characteristics, implementation details, and system integration can affect how data reaches processors.
The broader lesson is that data-center performance depends on balanced architecture. Faster accelerators cannot eliminate delays caused by insufficient storage throughput, network congestion, inefficient data placement, or poor software scheduling.
Cooling and Facility Design Still Matter
Power delivery and cooling are closely connected. Electrical energy consumed by computing equipment ultimately produces heat that must be removed from the facility. Higher rack power density can concentrate that heat in a smaller physical area.
Data-center designers therefore evaluate cooling systems, airflow, liquid-cooling options, heat exchangers, facility water systems, and operating conditions as part of the overall deployment.
A more efficient power converter can reduce some heat losses, but it does not remove the need to cool processors, memory, and other active components. Nor does a higher-voltage rack architecture guarantee a particular reduction in total facility energy use.
Reliable deployment requires integrated electrical and thermal planning, appropriate redundancy, maintainability, and careful validation under realistic operating conditions.
What the 800V Trend Means for Data-Center Operators
For operators, the decision to adopt a new power architecture should be based on workload requirements and total system economics. Relevant considerations include expected rack power, existing electrical infrastructure, equipment compatibility, conversion efficiency, cooling requirements, safety procedures, and maintenance skills.
A new architecture may be attractive for a greenfield facility designed around high-density AI systems. A retrofit of an existing data center can involve different trade-offs because the facility's electrical distribution, floor layout, protection systems, and operating procedures may already be established.
Before adopting a reference design, engineers should evaluate:
- Measured conversion efficiency across realistic operating loads.
- Electrical protection, insulation, grounding, and isolation requirements.
- Component compatibility and supply availability.
- Thermal behavior and cooling integration.
- Redundancy, serviceability, and fault-recovery procedures.
- Security configuration and firmware lifecycle management.
- Total cost of ownership compared with alternative architectures.
These evaluations help determine whether a new design provides practical value for a particular facility rather than relying on broad industry claims.
Does More Efficient Power Solve AI's Energy Challenge?
Power efficiency can improve the economics and environmental footprint of computing, but it does not by itself resolve the broader energy challenge associated with AI. Electricity demand depends on the amount of computing performed, the efficiency of the hardware, how systems are utilized, and the electricity supply serving the facility.
Better power conversion may reduce losses in one part of the system. Efficient software, workload scheduling, processor utilization, cooling, and facility design can contribute elsewhere. Renewable generation, grid capacity, energy storage, and local infrastructure also influence the overall picture.
There is a further distinction between efficiency and total consumption. If more efficient infrastructure makes AI computing cheaper, organizations may choose to run additional workloads. Total electricity demand can therefore rise even while energy efficiency improves at the equipment level.
For this reason, operators should measure both efficiency and absolute energy use. They should also consider the value of the computing performed, because reducing energy per task is more informative when paired with useful work completed.
What to Watch Next
The practical significance of the 800V trend will become clearer as reference designs, standards work, and deployed systems generate more evidence.
- Whether high-voltage DC architectures gain broader adoption in production AI racks.
- Measured efficiency and thermal results under representative workloads.
- How protection, safety, and maintenance procedures evolve.
- Compatibility across power components, server platforms, and rack designs.
- How operators balance electrical efficiency, cooling, networking, and facility costs.
The important measure is not whether a technology appears in a product announcement. It is whether it improves a complete deployment under the conditions in which that deployment will operate.
Frequently Asked Questions
Why are AI data centers considering 800V DC power?
Higher-voltage distribution can help manage current and resistive losses at high power levels, but the overall benefit depends on the complete electrical design.
What did Microchip and Navitas announce in October 2026?
They announced an 800V DC-to-6V DC reference design for AI data-center rack power applications using digital power control, GaN power devices, and hardware-based security.
What is gallium nitride used for?
GaN is a semiconductor material used in power electronics. Its switching characteristics can be useful in selected power-conversion designs.
Will 800V DC eliminate data-center cooling requirements?
No. It may help reduce some electrical losses, but processors and other components still generate heat that must be managed.
Should every data center adopt 800V power systems?
No. Operators should compare the architecture with their workload, facility constraints, safety requirements, compatibility needs, and total cost of ownership.
Conclusion
AI data centers require more than powerful accelerators. They need coordinated power delivery, efficient conversion, secure components, high-speed networking, storage, and cooling. The 800V DC reference design announced by Microchip and Navitas illustrates how the industry is exploring new approaches to these challenges. Its long-term value will depend on measured system performance, safe integration, and practical results in real deployments.