QuantumScape’s Solid-State PowerBlock Brings 4× Power Density to 1MW AI Data Center Racks

QuantumScape has taken its solid-state lithium-metal batteries out of the electric car and put them beside the AI server rack. On 8 October 2026, the San Jose company introduced the QS PowerBlock, a modular in-rack energy-storage reference design for AI data centers — the most concrete step yet toward making the solid-state battery data center combination a commercial reality.

The headline numbers are striking. QuantumScape says the system delivers four times the power density and five times the runtime of current data center industry standards, demonstrated on an expected AI training workload and benchmarked against the Open Compute Project’s Open Rack V3 (ORV3) specification. The adaptable design fits module, shelf and rack formats for industry-standard power rack configurations, and is aimed squarely at the 800VDC architectures that will feed the 1MW racks of next-generation data centers.

The solid-state battery data center pitch: 4× density, 5× runtime

Power density matters in a rack because batteries compete directly with compute hardware for the same scarce volume. Every litre given over to energy storage is a litre that cannot hold GPUs, networking or power-conversion equipment. QuantumScape’s argument is straightforward: if a physically smaller battery system can absorb the same power swings, more of the surrounding footprint can be devoted to useful compute — the metric AI builders now optimise for, alongside compute per megawatt and per cooling loop.

The runtime claim serves a different job. Longer runtime lets operators ride through short disturbances, smooth a bursty GPU load, bridge a temporary power shortfall, shave a peak-demand event, or buy more time for another source to respond. It is a notable reframing of AI data center energy storage: not emergency backup measured in minutes, but a working component of the power-delivery chain. In effect, the same hardware that protects the rack can help the whole facility behave as a more controllable grid load — an idea gaining ground as utilities wrestle with AI-driven demand growth. For background on how stationary storage earns its keep, see our battery energy storage explainer.

“With the QS PowerBlock, we are bringing this value proposition to AI data center customers, to enable the 800VDC designs that will power the next phase of AI growth,” said Dr. Siva Sivaram, chief executive of QuantumScape, in the company’s 8 October announcement.

Why 800VDC changes the battery equation

The industry’s shift toward 800VDC rack power is, at root, about current. For a given amount of power, raising the voltage lowers the current — and lower current means slimmer conductors, smaller resistive losses and far less copper threading through an already dense rack. QuantumScape has designed the PowerBlock as an 800VDC battery system from the start, instead of bolting a conventional low-voltage battery cabinet onto the side, pairing it with the power electronics already being developed for 800VDC distribution.

That design choice matters because the AI rack is becoming a single electrical machine: 800VDC distribution, power conversion, busbars, liquid cooling, protection and compute must all work together. A battery with strong cell-level numbers still has to fit inside that larger system, and designing for the target voltage from the start removes a layer of conversion hardware, cost and failure modes.

The safety margin that makes in-rack batteries thinkable

Putting more battery energy inside the rack creates an obvious tension: the storage must sit closer to the GPUs at exactly the moment the compute hardware becomes more expensive and the power density more extreme. Conventional lithium-ion installations manage that risk with chemistry choices, cell spacing, enclosures, monitoring, suppression and strict installation rules.

QuantumScape’s counter is a temperature margin. The company says its cells have demonstrated thermal stability well above 300°C in third-party and customer safety testing, compared with roughly 180°C — the point at which conventional lithium-ion cells typically enter catastrophic thermal runaway. If that margin holds in production systems, it directly addresses the fear of placing stored energy inches from hardware worth millions of dollars per rack.

The conceptual shift was put bluntly by the company’s data-center general manager Shahar Noy, who has described the AI-rack battery changing from a “spare tire” used only during failure into a “shock absorber” working continuously against GPU power swings, in remarks reported by BitcoinVersus’s analysis of the launch. Traditional UPS systems wait for the grid to disappear; the emerging model treats the battery as a buffer between a relatively steady grid and a highly dynamic compute load, smoothing voltage sags, transients and fast load steps before the GPUs ever see them.

Why it matters

The deeper story here is not a battery launch but a reclassification: rack-level energy storage is becoming first-class AI infrastructure. The old model treated batteries as insurance against losing electricity. The new model treats them as active equipment for shaping electricity — and if AI racks really do march toward 1MW each, drawing at times like a small industrial facility, that distinction becomes enormous.

There is also a commercial logic that favours the data center as solid-state’s first serious market. An EV battery must survive vibration, crash loads, weather, years of road cycling and brutal automotive qualification while staying cheap enough for mass-market cars. A data-center battery lives indoors in a controlled environment, is serviced by trained technicians, slots into standardised rack hardware, and can justify a premium price if it protects extremely valuable GPUs or lets a facility install more compute in the same building. Technologies still too expensive or too hard to qualify for millions of vehicles — including lithium-metal anodes, which promise far higher energy density than the LFP and NMC chemistries dominating EVs today — can clear the bar here first. QuantumScape is not alone in chasing solid-state scale; Gotion’s 400 Wh/kg solid-state cell shows how crowded that race has become.

Set against the numbers QuantumScape published, the comparison that counts is the one against the OCP Open Rack V3 benchmark: 4× the power density and 5× the runtime in the company’s demonstration, plus a thermal-stability margin of more than 120°C over conventional lithium-ion’s runaway point. Those are company claims from controlled testing, not independent measurements — but they define the bar the product will now be judged against.

Reference design, not a deployed fleet

The caveat that matters most is also the easiest to miss: the QS PowerBlock is a reference design. QuantumScape states plainly that its performance figures come from prototype testing and modelled system configurations under controlled conditions. Final performance will depend on product design, system configuration, manufacturing consistency, integration hardware, cooling, control software and real-world duty cycles. A demonstration is not a fleet of thousands of battery shelves running for years across production AI campuses.

QuantumScape has also spent years trying to scale its solid-state technology from laboratory cells to manufacturable products, and the familiar battery-industry question still applies: can the performance be produced repeatedly, safely and economically at scale? The data-center market changes the economics in the company’s favour, but it does not remove the manufacturing challenge.

The next public test comes quickly. QuantumScape has joined the Open Compute Project as a member, and the PowerBlock design will be on display at the 2026 OCP Global Summit, running 12–15 October at the San Jose Convention Center — booth C19, hosted by Megmeet, a power-system component provider in the NVIDIA MGX ecosystem. That venue is deliberate: the future AI rack will be decided as a complete system, and QuantumScape is claiming its seat at the table.