On 28 October, XPENG vice president Gu Jie said the solid-state battery developed for the IRON humanoid has entered small-batch production. It is the first such statement about a component the company has placed at the centre of the robot's design from the start. IRON itself, however, remains pre-SOP, meaning before the formal start of series production.
Key takeaways
- The battery was designed around the robot's limited internal space, not a car's floor pan
- The company stresses high energy density plus resistance to thermal propagation, impact and nail penetration
- No energy density, capacity, runtime, cycle life or production volumes have been disclosed
- The official IRON page advertises an all-solid-state battery, while earlier reporting described a semi-solid variant
- IRON production lines were commissioned in September, with market launch and deliveries targeted for 2027
Where this starts
A humanoid standing in a showroom has a few hours of shift ahead of it, handing out leaflets and greeting customers. The battery pack has to fit inside the torso, not along the floor beneath a cabin as in an electric car. Joint motors, control electronics and wiring sit right beside it.
If a cell fails, the heat has nowhere to escape. The robot stands among people, not in a car park.
But here is the problem: a high-energy-density battery in a tight enclosure is exactly the case where one damaged cell ignites its neighbours.
And that set of constraints is what the cell described by Gu Jie is meant to address — engineered not as a smaller car battery, but around the geometry and the risks of a walking machine.
What exactly did XPENG announce?
Yicai reported the remarks, and IT Home repeated the account, placing the appearance at the annual SAECCE conference. According to those accounts, the battery was designed specifically around the robot's limited internal space.
Gu described a high-energy-density design with measures to stop uncontrolled heating from spreading, plus resistance to impact and nail penetration. That is the standard safety test set for lithium cells.
What the announcement omits: energy density in watt-hours per kilogram, pack capacity, robot runtime, charge cycle life, production volumes and test conditions.
Without those figures there is no way to judge either IRON's real endurance across a shift or the scale of manufacturing. The announcement confirms entry into limited output rather than disclosing a rate or an independently verified safety level.
It is also worth understanding what small-batch production means in practice. It is the stage where a line makes enough cells for testing and the first product units, but without stable yield and without volume contracts. In batteries that is usually the hardest transition — from a working prototype to repeatable quality across thousands of units.
Why not just take a finished cell out of a car?
The best-mastered product of the battery industry today is the automotive pack. It lies flat in the floor, has room for cooling and barriers, and its maker holds years of durability data on it.
In a humanoid all three conditions change. The space is irregular and packed with actuators, the pack's mass loads the machine's balance, and the robot falls over — something a car does not do in normal use.
So the simpler approach, scaling an automotive cell down, only solves capacity. It solves neither the packaging, nor heat removal inside a tight torso, nor impact resistance in a fall.
There is a second problem, mentioned less often in press releases: liability. A machine working among shop customers needs documented resistance to thermal propagation, because without it the machine will not clear approval for public spaces. A manufacturer's declaration is the first step, not a certificate.
How is the competition solving this?
XPENG is betting on energy density, meaning longer work from a single pack. Agility Robotics took the opposite route and chose fast charging over a bigger battery.
In Digit 5 the new Nyobolt power system delivers 90 minutes of runtime and charges in 9 minutes. That is a 10:1 run-to-charge ratio, against 2:1 in the previous Digit 4 generation. The company converts that into more than 20 hours of productive work in a 24-hour day, which is more than two full shifts.
Both strategies attack the same limit — downtime. They differ on whether the bottleneck should be pack capacity or time spent at a charger. For a warehouse operator the second approach is easier to audit, because it needs no faith in a declared energy density: counting the minutes a robot stands idle is enough.
Early access to Digit 5 with the Nyobolt battery was announced for the first half of 2027, roughly the same window as IRON's market launch.
What do these names mean?
Four terms decide how much weight this announcement carries:
- All-solid-state — a cell whose electrolyte is entirely solid. There is no flammable liquid, so it is in theory safer and denser in energy, but harder to mass-produce.
- Semi-solid (solid-liquid hybrid) — the compromise: a partly solid electrolyte mixed with liquid. Easier to manufacture, without the full advantage of the fully solid variant.
- Pre-SOP — the stage before the formal start of series production. The equivalent of a release candidate in software: the line runs, but the product has not been frozen.
- Thermal propagation — overheating spreading from one cell to its neighbours. Blocking that effect is a baseline safety requirement for any pack.
The difference between the first two bullets decides whether the announcement is a manufacturing breakthrough or one more step inside a known technology.
Is it working yet?
At component level — small-batch production. At robot level — still before the start of production. Gu described IRON as pre-SOP, which is consistent with separating the first units off a line from the formal launch of series manufacturing.
XPENG commissioned the IRON production lines in September and showed a completed robot walking off the line. That announcement still targeted mass production by the end of 2026 plus market launch and deliveries in 2027.
The battery is only one part of preparing for commercial use. The company is also lining up its supply chain and has previously shown interaction demos — handshakes, high-fives and fist bumps — matching its plan to enter stores and other service settings first.
Why it matters
Range on a single charge is today the hard limit on how useful humanoids are in shift work. A battery announcement therefore matters more than another gesture video. At the same time, the missing figures and the unresolved electrolyte show the industry's pattern: a manufacturing milestone gets announced faster than the data needed to check it.
What's next
- IRON mass production targeted for the end of 2026, with market launch and deliveries in 2027 per the company's September announcement
- European market entry planned for the second half of 2027
- The electrolyte stays an open question — until the maker describes the cell construction, the all-solid-state versus semi-solid discrepancy remains unsettled
Sources
- Humanoids Daily — XPENG Says IRON's Solid-State Battery Has Entered Small-Batch Production
- CnEVPost — XPeng's IRON humanoid robot shifts to semi-solid-state batteries
- XPENG — AI Robot IRON
- Humanoid Robotics Technology — Agility Robotics Partners with Nyobolt to Power Digit 5 Humanoid Robot





