Solid-State Battery for AI Racks Explained: QuantumScape PowerBlock's 4x Power Density and 300 °C Safety Claim

 

Bottom line: the first big paying market for solid-state batteries may be AI racks, not EVs. On October 8, QuantumScape (QS) launched its first data center product, the QS PowerBlock, pitching power density, runtime and fire safety in one in-rack package.

What happened

• Press releaseQS PowerBlock is a modular reference design that can be built as modules, shelves or full racks in industry-standard power rack formats, supports 800 VDC architectures, and targets next-generation 1 MW racks.
• Press releaseIn a demo running an expected AI training workload, QS claims 4x the power density and 5x the runtime of the OCP Open Rack V3 (ORV3) specification.
• Press releaseQS cells showed thermal stability well over 300 °C in third-party and customer safety tests; QS says conventional lithium-ion cells typically enter catastrophic thermal runaway at about 180 °C.
• Press releaseQS states the results come from prototype testing and modeled system configurations. The design will be shown at the OCP Global Summit, Oct 12–15, at the Megmeet booth.
• InterviewCEO Siva Sivaram told Axios that customer talks are underway and large training data centers will likely come first.

The baseline: Confirmed, OCP specAn ORV3 battery backup unit (BBU) shelf holds six 3 kW modules (5+1 redundancy) and must deliver 15 kW for more than 240 seconds. That is at least about 1 kWh per shelf (15 kW × 240 s, our arithmetic). This is the yardstick behind the 4x and 5x claims.

Engineering point 1: why a rack needs a buffer battery

A classic UPS sits in a separate room and bridges an outage until generators start. AI training is a different load: thousands of GPUs compute and wait in lockstep, so rack power can swing sharply over short intervals. Putting storage inside the rack lets it absorb those swings next to the GPUs instead of pushing them upstream into the distribution system and the grid. QS frames the extra runtime as more margin "to smooth out power fluctuations." That is also why the industry is looking at supercapacitors here: very fast and long-lived, but low in stored energy.

Engineering point 2: what 4x power density means in hardware

Our estimate, assumed valuesTake a 1 MW rack and apply the ORV3 idea of 240 seconds of ride-through. That is about 67 kWh (1,000 kW × 240 s). Built from 15 kW ORV3 shelves, you would need the discharge power of roughly 67 shelves. If the 4x power density holds, the same power fits in about a quarter of the volume, around 17 shelf-equivalents. This is a scaling illustration only; QS has not published PowerBlock dimensions or capacity.

Engineering point 3: 5x runtime, under what conditions?

Our estimate, assumed valuesIf the 5x figure was measured at the same 15 kW load, runtime goes from 240 seconds to about 1,200 seconds (20 minutes), or roughly 5 kWh. But the release does not state the baseline condition, or whether the comparison is per volume or per power level. A lithium-metal anode helps energy density; delivering high power for long durations also depends on cell resistance, heat removal and depth of discharge, none of which have been disclosed.

Engineering point 4: 300 °C versus 180 °C

Press release figuresThe gap is more than 120 °C (our arithmetic). In conventional cells, flammable liquid electrolyte and separator failure drive much of the runaway risk; QS pairs a ceramic solid-state separator with a lithium-metal anode, which should tolerate heat better in principle. Sivaram even argues a safer battery can ease insurance and permitting for a data center. As we covered in Who Says Solid-State Batteries Won't Catch Fire?, solid-state does not mean fireproof: lithium metal is reactive, and dendrites and mechanical stress still matter. So far there is only a statement about third-party tests, with no public method or data.

Open questions

• The baseline and test conditions behind 4x and 5x (same volume? same power? which load profile?) are not public.
• No cell-level energy density, power capability or cycle life. Power smoothing means many shallow cycles, so cycle life matters a lot.
• The 300 °C test method and the third-party lab are not named.
• Lithium-metal cells usually need some stack pressure; how the module handles it, and at what cost, is not described.
• No pricing, production timeline or named customer.

Watch on The Engineering Core YouTube channel

• Solid-State Batteries Just Killed Supercapacitors at CES 2026?
• Who Says Solid-State Batteries Won't Catch Fire? The Hidden Engineering Nightmare
• AI's Power Crisis: SOFC vs SMR Nuclear — The Ultimate Energy Showdown!
• The Ultimate EV Battery Guide: NMC vs. LFP Explained

Your turn: if you were designing the buffer storage next to a GPU rack, what would you pick: solid-state lithium-metal, supercapacitors, or conventional lithium-ion with heavier fire protection? Tell me in the comments.

Sources: QuantumScape press release (Oct 8, 2026), Axios interview (Oct 8, 2026), OCP Open Rack V3 BBU Shelf spec Rev 1.1. The 1 kWh, 67 kWh, 67/17 shelf-equivalent, 20 min/5 kWh and 120 °C figures are our own arithmetic from published numbers and assumed values, not QS data.

The Automation Core | theautomationcore.blogspot.com | The Engineering Core YouTube channel