Navitas has unveiled a 2,300-volt silicon carbide (SiC) power module featuring an 8 mΩ on-resistance, targeting both utility power grids and AI data center power delivery. The high-voltage SiC device is designed to support the industry's shift toward 800V HVDC architectures and higher-density power conversion. It addresses a less-discussed but critical layer of AI infrastructure beyond GPUs, memory and networking.
Overview
On October 21, 2026, Navitas Semiconductor introduced the G4H08MT23GB4, a new family of silicon carbide power modules rated for voltages up to 2,300 volts and specified at an on-resistance of 8 milliohms. The company positions the devices for solid-state transformers, energy storage systems, and high-power industrial converters, with samples already available. The announcement arrives at a moment when discussion of AI infrastructure is dominated by GPUs, memory, and network bandwidth, while the electrical layer that feeds those systems receives far less attention.
Device Architecture and Ratings
The module is built on Navitas's fourth-generation GeneSiC technology, using what the company calls a Trench-Assisted Planar (TAP) architecture. TAP combines elements of planar MOSFET structures with an additional trench architecture intended to improve the electrical properties of the power semiconductor. The device is offered in a full-bridge topology inside a SiCPAK G package.
Silicon carbide is a wide-bandgap material. Compared with conventional silicon, it tolerates higher electric field strengths, which makes it attractive for devices with high voltage ratings. In high-power converters, that translates into the potential for reduced conduction and switching losses with appropriate circuit design, while higher switching frequencies can allow smaller passive components. Navitas describes the 8 mΩ figure as particularly low within the relevant SiCPAK class, though the company's own framing carries an important qualification.
Why On-Resistance Is Not Efficiency
On-resistance does not represent total power dissipation. Beyond conduction losses, switching frequency, current, temperature, and gate drive all influence efficiency. A direct efficiency comparison with competing products cannot be inferred from the resistance specification alone. The same caution applies to the module's headline voltage rating: 2,300 volts defines a capability envelope, not a measured system-level outcome.
Packaging, Substrate, and Thermal Design
Navitas has revised the mechanical and thermal construction alongside the semiconductor itself. The module uses an AMB (Active Metal Brazing) substrate based on aluminum nitride — a joining process that attaches metal structures to ceramic substrates. Aluminum nitride offers high thermal conductivity while retaining good electrical insulation, and Navitas specifies an insulation rating above 5 kV.
The encapsulant has also changed: epoxy resin replaces conventional silicone gel. According to the manufacturer, this improves thermal conductivity by a factor of ten compared with silicone gel, and is also intended to reduce mechanical stress and the ingress of moisture and contaminants. Independent measurements of these improvements are not yet available.
Variants, Connections, and Qualification
Four variants are offered. Alongside the standard G4H08MT23GB4, a -T version ships with pre-applied thermal interface material. Variants carrying the HC designation use connections designed for higher current loads, and are also available with thermal interface material. For those High-Current parts, Navitas uses reinforced press-fit connections whose current-carrying capacity per pin is said to be twice that of the standard connections.
Navitas has announced qualification of the module series under its internal AEC-Plus program, which it says goes beyond certain AEC-Q101 and JEDEC requirements. Planned tests include additional evaluations under high electrical voltage, temperature stress, and dynamic switching conditions, along with extended temperature cycling and power cycling. The wording matters here: AEC-Plus is a manufacturer-specific standard, and the announcement should not be equated with a fully completed independent certification. For development work, Navitas offers a double-pulse test system on request, allowing switching events and resulting electrical losses to be examined under defined conditions. Pricing for volume production and a specific mass-production start date have not been announced.
Market Context: The Power Layer of AI Infrastructure
The applications list is broad: solid-state transformers, large battery storage systems, solar inverters, industrial drives, and megawatt charging systems for electric vehicles. That breadth indicates the module was not developed exclusively for data centers but is generally aimed at high-power energy conversion. Solid-state transformers are a particularly notable target — unlike conventional transformers, they use power semiconductors and electronic circuits to convert and regulate electrical energy, and could serve future power supply architectures where large amounts of power must be efficiently converted between DC voltage levels.
Implications for AI Data Centers
Navitas explicitly points to AI data centers, where such modules could form part of the power supply between the grid connection and the high-voltage DC busbars in server racks. This maps onto the industry's movement toward 800 V HVDC architectures and higher-density power conversion. The significance is structural rather than incremental: as rack power densities climb, the losses and thermal burden of conversion stages become a constraint on deployment, not an afterthought. A 2,300-volt SiC module with low on-resistance and a redesigned thermal stack is a plausible building block for that layer, even if the data center is only one of several target markets.
Future Outlook
The technical case for the GeneSiC G4H08MT23GB4 is coherent: high blocking voltage, low specified on-resistance, and a reworked substrate and encapsulant aimed squarely at thermal and insulation performance. What is missing is external validation. Only independent measurements under realistic operating conditions will show how significant the advantage over existing SiC modules actually is, and the absence of volume pricing and a production timeline leaves the commercial picture open. For now, the module is best read as a signal — that the development of AI hardware depends on more than faster processors, and that power electronics must keep pace with the requirements those processors impose.