AI data centers are racing toward power levels that look less like server rooms and more like industrial plants, and their electrical infrastructure is straining to keep up. Into that crunch steps an 800V immersion-cooled battery platform that borrows tricks from electric vehicles to deliver dense, fast, and thermally stable backup power right where GPUs need it most.
By pairing high-voltage DC with liquid immersion cooling, this new battery architecture aims to shrink copper, tame heat, and smooth brutal load swings from AI training clusters. It is also arriving just as the broader ecosystem, from power electronics to cooling vendors, pivots around the same 800V benchmark, signaling that backup batteries are about to become a central design lever for next-generation AI facilities rather than a bolt-on afterthought.
Why AI data centers are converging on 800V
AI workloads are pushing data centers from megawatt to gigawatt campuses, and at those scales, low-voltage distribution wastes too much energy and metal. Operating at higher voltage levels cuts current for the same power, which in turn reduces resistive losses and shrinks the size of busbars and cables. One analysis notes that Operating at elevated DC levels can reduce the amount of copper required by up to 45%, a material and space saving that matters when every rack is packed with accelerators.
That logic is driving a shift toward 800V high-voltage DC architectures across the power chain, from grid interfaces to rack-level converters. Technical voices such as Jeff Morroni frame 800V high-voltage DC architectures as a direct response to rapid growth in the server and artificial intelligence markets, where efficiency, sensing, protection, and safety isolation all have to scale together. In parallel, industry roadmaps describe how the transition to an 800V high-voltage DC backbone is becoming a prerequisite for the power density demanded by modern AI workloads, a point underscored in analyses of Pioneering 800V HVDC Power Distribution for Next Generation AI Data Centers.
The 800V immersion-cooled battery that wants to live in your rack
Into this 800V ecosystem steps a new immersion-cooled battery module designed specifically for AI backup power rather than repurposed from legacy UPS gear. The unit is built as a compact building block that measures 2 OU in height and supports voltage configurations of ±400V or 800V, so a standard 20 OU rack can deliver megawatt-scale energy within a compact rack footprint. That form factor and voltage flexibility are central to the design described in reports on the module, which is explicitly aimed at power-hungry AI data centers.
The system comes from XING, a company that has spent years refining immersion-cooled packs for electric vehicles and energy storage before turning that experience toward data center backup. The 800V immersion-cooled battery is Designed for megawatt-scale racks, with automotive-grade validation intended to keep system stability under extreme loads when AI clusters ramp from idle to full power in milliseconds. By embedding the battery directly into the rack envelope, the design aims to shorten power paths, reduce conversion steps, and make backup power feel less like a distant insurance policy and more like an integrated part of the compute stack.
XING Mobility’s decade-long immersion experiment pays off
The battery did not appear in a vacuum. XING has spent a decade iterating on immersion-cooled packs across electric vehicles, stationary storage, and now AI backup, and it is using CES as a stage to show how those threads converge. In Taipei, XING Mobility Showcases a Decade of Immersion Cooled Battery Innovation at CES, highlighting three major applications across EV, ESS, and AI Data Center BBU that share a common liquid-cooled core. That history matters because immersion cooling at pack level is still relatively exotic in mainstream data center design, and operators want proof that the chemistry and mechanics can survive real-world abuse.
In its AI-focused push, XING Mobility is positioning the backup battery unit as one of those three flagship applications, alongside automotive and grid-scale systems. The company has framed this as part of a broader Decade of Immersion Cooled Battery Innovation, using CES to argue that the same thermal tricks that keep high-performance EV packs stable at full throttle can deliver high efficiency across diverse applications. That narrative is reinforced in a separate announcement where XING Mobility Showcases a Decade of Immersion Cooled Battery Innovation at CES and explicitly calls out the AI Data Center BBU as a peer to its EV and ESS products, not a side project.
Inside the BBx800: a backup battery built for AI spikes
At the heart of XING’s data center play is the BBx800, an 800V immersion-cooled backup battery unit that tries to solve a very specific problem: AI clusters do not draw power smoothly. Training runs and inference bursts can slam power rails with rapid transient demand fluctuations that traditional UPS systems struggle to follow without oversizing. XING Mobility launched the BBx800 to target exactly that AI data center power sector, pairing immersion cooling with an 800V DC interface so the pack can respond quickly without cooking itself.
In its CES messaging, XING Mobility will showcase the world’s first immersion-cooled 800V high-voltage DC BBU alongside Project V, an all-electric concept that underscores how much of its technology is shared between mobility and data center use cases. The company stresses that, However, 800V systems introduce significant technical complexity, including requirements for advanced insulation design and the ability to handle rapid transient demand fluctuations without compromising safety. By bringing an immersion-cooled 800V high-voltage DC BBU and Project V to the same stage, XING is effectively arguing that if its packs can survive track-day EV abuse, they can also ride out the violent load steps of next-generation AI data centers.
Why immersion cooling and 800V belong together
Running batteries and power electronics at 800V concentrates a lot of energy in a small volume, which makes thermal management a first-order design constraint rather than a late-stage patch. Immersion cooling addresses that by submerging cells and busbars in a dielectric fluid that can pull heat away uniformly, avoiding the hot spots that plague air-cooled packs. XING Mobility Showcases this approach as a Decade of Immersion Cooled Battery Innovation, arguing that immersion enables higher power density and more consistent cell temperatures across EV, ESS, and AI Data Center BBU applications, as highlighted in its Mobility Showcases narrative.
The broader data center ecosystem is also moving toward liquid cooling as rack power climbs, which makes immersion-cooled batteries feel less alien than they might have a few years ago. Vertiv’s $1B PurgeRite acquisition marks a bold move towards advanced liquid cooling and thermal management solutions for AI-heavy data centers, a signal that large infrastructure vendors see fluid-based systems as the default for future high-density halls. That context matters for XING’s strategy, because a facility that is already plumbing racks for liquid-cooled GPUs is more likely to accept an immersion-cooled BBU as part of its thermal stack, especially when companies like Vertiv are already betting heavily on liquid infrastructure.
800 VDC is becoming the backbone of AI power plants
While XING focuses on backup, other players are rebuilding the main power spine of AI facilities around 800 VDC. Delta has been particularly vocal, presenting its Groundbreaking 800 VDC Power Solutions Showcased at OCP Global Summit to enable what it calls sustainable AI factories. In that vision, Delta‘s Groundbreaking 800 VDC Power Solutions Showcased at OCP are pitched as a way to meet the demands of hyperscale AI computing while trimming conversion losses and simplifying distribution.
Industry presentations such as Introducing Next-Gen 800 VDC Data Centers describe how, as data centers scale from megawatts to gigawatts in the AI era, a new power architecture is essential for efficiency and controllability. In that framing, 800 VDC is not just a number but a system boundary that lets designers standardize converters, busways, and protection schemes from the substation to the rack. The Oct discussion of next-gen 800 VDC data centers reinforces that the industry is coalescing around this voltage as a sweet spot between efficiency, safety, and component availability, which in turn makes it natural for backup batteries like the BBx800 to speak the same language.
Wide bandgap semiconductors make 800V practical
None of this 800V enthusiasm would be viable without a parallel revolution in power semiconductors. Traditional silicon devices struggle with the switching speeds and efficiency needed at high voltage and high frequency, especially in compact data center gear. Companies like Navitas are leaning on Wide Bandgap materials, specifically Gallium Nitride and Silicon Carbide, to build converters that can handle 800V with lower losses and smaller magnetics. By leveraging Wide Bandgap (WBG) materials such as Gallium Nitride and Silicon Carbide, Navitas argues it can reshape how electricity is managed at scale for both AI and mobility.
For data centers, that means rectifiers, DC-DC stages, and battery interfaces that can run closer to 800V without ballooning in size or wasting energy as heat. It also enables more granular control of power flows, which is critical when pairing fast-responding BBUs with volatile AI loads. As WBG devices mature, I expect to see more integrated solutions where an 800V immersion-cooled battery, a Gallium Nitride based converter, and a liquid-cooled GPU rack are treated as a single engineered system rather than separate procurement lines, tightening the coupling between storage, conversion, and compute.
From grid to rack: stitching BBUs into 800V HVDC networks
The real test for XING’s 800V immersion-cooled battery will be how cleanly it plugs into emerging HVDC distribution schemes. Analyses of HVDC Power Distribution for Next Generation AI Data Centers describe a future where an 800V DC bus runs deep into the white space, feeding racks directly instead of stepping down to multiple AC tiers. In that topology, a rack-level BBU that already operates at 800V can sit natively on the bus, acting as both a buffer and a local energy reservoir without extra conversion stages.
That integration also changes how operators think about resilience. Instead of a monolithic UPS room feeding an entire hall, each row or rack can carry its own 800V BBU, coordinated by software to ride through faults, shave peaks, or even participate in grid services. Given the complexity of 800V high-voltage DC architectures, including the need for precise sensing, protection, and safety isolation highlighted by Given the rapid growth in AI markets, I see BBUs like the BBx800 as both a technical and an organizational bridge between traditional UPS thinking and a more distributed, software-defined power layer.
The partnership and ecosystem bets behind XING’s move
XING knows it cannot reshape data center power alone, so it is pairing its battery expertise with established power electronics and OEM partners. The company has already teamed up with Acbel Polytech to explore how the BBx800 can be integrated into complete power systems for AI data centers, combining Acbel Polytech’s experience in power supplies with XING Mobility’s immersion-cooled packs. That kind of partnership is essential if the BBU is to show up not as a bespoke add-on but as a pre-engineered option in server and rack configurations.
At the same time, the broader ecosystem is aligning around 800V, from Delta’s Groundbreaking 800 VDC Power Solutions Showcased at OCP to Vertiv’s $1B PurgeRite acquisition for liquid cooling. When I look across these moves, I see XING’s 800V immersion-cooled BBU as a missing puzzle piece that connects high-voltage distribution and advanced thermal management directly to the battery layer. If AI data centers are the new power plants of the digital economy, then an 800V immersion-cooled battery sitting quietly in each rack may soon be as fundamental as the GPUs it protects.
More from MorningOverview