By Shaun Harris
VP, Solutions Technology
[3 minutes]
Coming from a family of pilots and machinists, I have often wondered what drove the Wright brothers to step away from their thriving Wright Cycle Company to pursue the uncharted territory of heavier-than-air flight. At the time, bicycles were a booming and reliable form of transportation, while automobiles and motorcycles represented the next logical steps for the internal combustion engine. For some reason, the Wright boys decided to put an engine into something that could fly (they hoped). Throughout engineering history, we see moments of "marked change"—pivots away from conventional, tried-and-true methods and the obvious next step toward something fundamentally better.
I believe the shift to an 800 VDC power architecture is one of those historic inflection points.
High Current Is Crowding Out Compute
Engineering breakthroughs are rarely serendipitous; they are typically forced by performance barriers and unacceptable compromises. When a single component or standard prevents an entire system from scaling, it must change.
In the AI/ML landscape, the defining metric is performance per watt—maximizing compute while minimizing the energy required to achieve it. Today, that drive is constrained by the sheer power demands of modern GPU architectures.
As GPU power requirements increase exponentially, rack density climbs alongside them. Because the limits of copper signal integrity require GPUs within a rack to remain in close physical proximity, squeezing more compute into a tight physical volume in turn drives current connections to extreme levels. High current demands massive copper connectors and bulky high-current cabling, which quickly commandeer all the rack’s physical space. The only practical way to break this bottleneck—and enable higher-power GPUs within the same footprint—is to increase the delivery voltage to the blade, thereby dramatically reducing current and connector size.
800 VDC Is the Best Standard for Today
Is 800 VDC the optimal target?
I was skeptical initially and argued that 800 VDC wasn't necessarily the ideal voltage level. However, after evaluating extensive modeling across alternative architectures—including ±400, 740, 900, and 3000 VDC—the performance data changed my mind. While each approach has minor trade-offs, I believe 800 VDC’s alignment with the broader commercial ecosystem (such as the automotive EV supply chain) makes it the most practical, immediate path forward.
Caveat: Future mega-clusters may demand even higher voltages, but for our current era, 800 VDC is the right standard.
The Awesome Benefits of SSTs
Delta Electronics showcased its medium-voltage (MV) 800 VDC solid-state transformer (SST) platform, designed specifically to tackle grid-to-chip power conversion for high-density AI clusters, at the 2025 Open Compute Project (OCP) Global Summit. Delta has been pioneering MV to low-voltage (LV) 800 VDC SSTs for nearly a decade.
The MV-to-800 VDC SST architecture dramatically reduces complexity:
- Fewer transformers: Eliminates traditional MV to 480 VAC and 480 to 415 VAC step-down transformers.
- Streamlined conversion: Reduces power conversion stages from four to two compared to traditional 480 VAC to 12 VDC lineups.
- Higher efficiency and smaller footprint: Yields direct energy-efficiency gains and frees up significant floor space.
800 VDC Scales Past 1 MW per Rack
Understandably, much of the data center industry has approached 800 VDC with caution, often preferring to keep standard 480 VAC distribution to the row. Operators cannot risk stranding assets or making costly retrofits with every new hardware generation when they’re committing $1+ billion to a facility with a 25-year structural lifecycle and housing GPUs that need to be refreshed every 5 to 6 years.
Fortunately, analyses demonstrate that 800 VDC distribution can comfortably scale to support rack densities exceeding 1 MW—ensuring such facility infrastructure will stay future-proof across multiple GPU roadmap cycles. And we are now seeing the data center industry shift beyond proof-of-concept evaluation and testing into commercial reality, with data center RFQs specifying MV SST power delivery for 250 to 500 MW builds.
Take the Next Step: Liftoff!
Just as the Wright brothers recognized when it was time to leave the pavement behind and take to the air, the industry is recognizing that 800 VDC SST is the key to elevating high-density compute in the next era. So please fasten your seatbelt. We are cleared for takeoff.
Learn more about Delta’s SST platform.
Q&A
Q: Why is the data center industry moving to 800 VDC power architecture?
A: Raising the delivery voltage to 800 VDC dramatically reduces current and connector size, enabling higher-power GPUs within the same footprint. 800 VDC breaks the power bottleneck that limits GPU density in AI data centers. As GPU power requirements climb, rack density rises with them, and copper signal-integrity limits force GPUs within a rack to stay physically close. Delivering all that power at lower voltages drives current to extreme levels, and the massive copper connectors and bulky cabling that high current demands quickly consume the rack's physical space.
Q: Why 800 VDC instead of another voltage level?
A: 800 VDC aligns with the broader commercial ecosystem, which makes it the most practical, immediate path forward. Modeling across alternative architectures, including ±400, 740, 900, and 3000 VDC, showed that each option carries minor trade-offs. 800 VDC, however, taps into existing supply chains such as the automotive EV market. Future mega-clusters may eventually demand even higher voltages, but 800 VDC is the right standard for the current era.
Q: What is Delta's MV-to-800 VDC solid-state transformer (SST) platform?
A: Delta's MV-to-800 VDC SST platform manages grid-to-chip power conversion for high-density AI clusters. Delta showcased the platform at the 2025 Open Compute Project (OCP) Global Summit, building on nearly a decade of pioneering work in medium-voltage (MV) to low-voltage (LV) 800 VDC SSTs.
Q: How does an MV-to-800 VDC SST architecture simplify data center power delivery?
A: The SST architecture cuts both equipment and conversion stages, delivering efficiency gains and a smaller footprint. Specifically, it:
- Eliminates traditional MV-to-480 VAC and 480-to-415 VAC step-down transformers
- Reduces power conversion stages from four to two compared to traditional 480 VAC to 12 VDC lineups
- Yields direct energy-efficiency gains and frees up significant floor space
Q: Why have some data center operators been cautious about adopting 800 VDC?
A: Operators worry about stranded assets and costly retrofits across GPU generations. A new facility often represents a $1+ billion commitment with a 25-year structural lifecycle, yet it houses GPUs refreshed every 5 to 6 years. That mismatch has led many operators to keep standard 480 VAC distribution to the row.
Q: Is 800 VDC infrastructure future-proof, and is it ready for commercial deployment?
A: Yes. 800 VDC distribution comfortably scales to support rack densities exceeding 1 MW, keeping facility infrastructure viable across multiple GPU roadmap cycles. The industry has also moved beyond proof-of-concept testing into commercial reality: data center RFQs now specify MV SST power delivery for 250 to 500 MW builds.