The company is also presenting two solid-state transformer topologies for converting medium-voltage grid power directly to 800 V HV DC
Decision Focus
On May 18, 2026, Navitas Semiconductor announced it will demonstrate GaN and SiC power solutions for AI data centers at PCIM Europe in Nuremberg, June 9–11. The demonstrations include an 800 VDC power delivery board that removes the traditional 48V intermediate bus converter stage from the rack power chain. For Global Heads of Data Center Energy, the operational signal is not the product launch — it is the confirmation that the industry is actively engineering around a rack-level voltage architecture that most infrastructure planning cycles have not yet absorbed.
90-Second Brief
Now, navitas will show a 20 kW 800 V-to-6 V power delivery board targeting 97.5% peak efficiency and a 10 kW 800 V-to-50 V DC-DC platform rated at 98.5% peak efficiency and 2.1 kW per cubic inch. Both platforms are built for 800 VDC AI compute environments and are aligned with NVIDIA’s 800 VDC AI factory architecture, which Navitas first supported in October 2025. The company is also presenting two solid-state transformer topologies for converting medium-voltage grid power directly to 800 V HV DC. Suppliers are now demonstrating production-grade hardware rather than laboratory prototypes, a stage that, by typical industry timing, precedes hyperscaler infrastructure commitment by 12 to 24 months.
What Is Really Happening?
The 48V intermediate bus converter has been the standard workhorse of data center rack power delivery for years. It steps voltage down from the facility distribution level to 48V, then again to the processor supply rail. That two-stage conversion path carries real efficiency losses at the scale of a hyperscale AI cluster. The 800 VDC architecture collapses that chain: power arrives at the rack at 800 V and converts directly to the processor voltage in a single stage. The Navitas press release states the 20 kW demonstration board explicitly targets elimination of the IBC stage while improving system efficiency, reliability, and power density simultaneously.
The SiC side of the demonstration is equally relevant for data center energy operators. Navitas is showcasing two solid-state transformer topologies — using 3300 V and 1200 V SiC devices — capable of converting medium-voltage grid power directly to 800 V HV DC. The lineup includes an EPFL-developed full SST cell and a 50 kVA bi-directional active front end DAB SST solution using 3300 V SiCPAK MOSFET modules. Together, these represent a pathway where the substation-to-rack conversion chain becomes materially shorter. This is not an incremental refinement to existing power topology — it is a structural redesign of how grid energy reaches compute hardware.
Why It Matters for Global Heads of Data Center Energy
The efficiency figures carry direct budget implications at portfolio scale. Conversion stages operating at 97–98.5% peak efficiency rather than the 93–95% range typical of legacy two-stage architectures reduce heat dissipation per rack. Lower heat means reduced cooling load, flowing directly into PUE and Scope 2 emissions calculations. For operators managing multi-GW portfolios with board-level sustainability commitments, even a one-point PUE improvement across a large campus is material.
The harder infrastructure question is sequencing. Facilities designed around 48V distribution buses require different PDU architecture, different cabling specifications, and different UPS configurations than an 800 VDC facility. If hyperscalers commit to 800 VDC rack standards for next-generation AI compute — and NVIDIA’s AI factory platform already points in that direction — then data centers ordered today may be built to a standard that is one compute generation behind. That creates a stranded-capacity risk that is not about power availability but about power delivery compatibility.
The solid-state transformer demonstrations introduce a second planning dimension. SST technology capable of converting directly from medium-voltage grid to 800 V DC could reduce the number of transformation steps between the utility interconnect and the compute load. Fewer conversion stages mean simpler power infrastructure, potentially smaller substation footprints, and shorter paths through transformer supply chains that are already constrained by two-to-three year lead times. Whether this SST pathway reaches commercial scale within a relevant planning window is not yet confirmed, but the hardware is now being demonstrated at 50 kVA by a company with production manufacturing relationships.
Forward View
If 800 VDC becomes the dominant rack power standard for AI compute by 2027 or 2028, three fronts are worth tracking. First, PDU and UPS vendors will need to certify 800 VDC-native products; the qualification timeline for data center power infrastructure components runs 18 to 24 months ahead of deployment. Second, utility interconnection and substation design specifications may need revision if solid-state transformers become viable at scale — the interface between grid and data center changes if the SST replaces the step-down transformer stack. Third, PPA and tariff structures that assume a given PUE baseline for energy cost allocation will need updating if facility-level efficiency improves materially through architecture change rather than cooling optimization.
What Is Still Uncertain
The Navitas demonstrations at PCIM are manufacturer specifications, not audited operational benchmarks from deployed data center environments. The claimed efficiency figures represent peak performance under controlled conditions; real-world derating under variable load profiles, thermal conditions, and duty cycles has not been independently verified in this context. There is also no confirmed deployment timeline from a hyperscaler or colocation operator for 800 VDC infrastructure at scale. The NVIDIA AI factory alignment is a directional signal, not a binding infrastructure commitment by operators.
The solid-state transformer pathway carries additional uncertainty. SST technology for utility-scale power conversion is earlier in its commercial maturity than rack-level power delivery hardware, and the 50 kVA demonstration is several orders of magnitude below what a large data center substation requires. Whether the technology scales economically and within relevant planning timelines remains an open question.
One Question for Your Team
At what point in the facility design cycle does a commitment to 48V distribution architecture become a structural constraint against 800 VDC AI compute deployment — and does any active site in your pipeline cross that threshold before the end of 2027?
Sources
- Stocktitan — Navitas GaN, SiC power for AI data centers at PCIM | NVTS Stock News (Link)
