Why Pouch Cells Cant Power AI Glasses Inside the Ultra-Narrow Steel-Can Battery Deep Dive

The Form Factor Problem Nobody Talks About

When you're designing AI glasses, the battery is the constraint that dictates everything else. The temple arm of a pair of smart glasses is narrower than an adult's pinky finger, yet it has to power a camera, speakers, an AI inference workload, and in some models, a display.

Pouch cells — the soft, foil-wrapped batteries in every phone and laptop — simply can't do this job. Their folds waste volume. Their tolerances eat millimeters. And at very small sizes, they struggle to deliver peak power when multiple subsystems fire at once — like when a user is recording video and asking the AI a question simultaneously.

The engineering answer is a battery that's rigid, precise, and shaped to the product instead of the other way around: the steel-can cell.

This analysis is based on Meta's engineering writeup — see the original engineering deep dive for the full context.

Steel-Can Cells at 7mm: Rethinking Every Internal Component

Steel-can batteries aren't new — power tools and watches have used them for years. The hard part was scaling them down to 7mm widths, narrower than anything that previously existed. That required reinventing the battery's internals.

Electrode Architecture: From Jelly Roll to Stacked Layers

Traditional steel-can cells use a wound "jelly roll" of electrode material. The Meta team replaced this with die-cut stacked layers, effectively wiring many small cells in parallel.

Traditional (Jelly Roll):        Meta (Stacked Layers):

  ┌─────────────┐                  ┌──┬──┬──┬──┐
  │  ◎ wound    │                  │  │  │  │  │
  │  electrode  │      →           ├──┼──┼──┼──┤
  │  spiral     │                  │  │  │  │  │
  └─────────────┘                  └──┴──┴──┴──┘
  High impedance                   Low impedance
  (peak power limited)             (peak power capable)

The payoff is dramatically lower impedance. That matters because low impedance prevents brownouts when the device demands a power spike — for example, video recording plus an AI inference request hitting the SoC at the same time.

Tolerances: 100 Microns Is the Whole Game

A steel-can cell holds its shape to roughly 100 microns. On a 10mm-wide battery, that tolerance reclaims usable volume that translates directly into energy density and runtime. Pouch cells, by contrast, deform and waste that space.

Capacity Growth Across Generations

Notice the Gen 1 → Gen 2 jump: cell capacity only grew ~30%, but runtime claims doubled. The chemistry didn't change — the gains came from system-level efficiency: tighter power management, firmware control, and a form factor that allowed a larger cell to fit.

The Dual-Battery Puzzle and What to Watch For

The Cross-Charging Problem

The Oakley Meta Vanguards put a battery in each temple arm. That sounds simple, but it creates a real systems problem at the intersection of electrical, firmware, and mechanical engineering.

The cells are symmetric — but the electrical loads are not split evenly between the two sides. One side might drive the camera and SoC, the other might drive speakers. That asymmetry creates:

• Cross-charging risk — one cell charging the other through the shared bus
• Boot/shutdown sequencing complexity — both cells must agree on state during power transitions
• Firmware coordination — the power management IC has to arbitrate which cell serves which load

If you're building a dual-battery wearable, budget serious engineering time for the boot sequence state machine. It's where most teams underestimate complexity.

Critical Caveats

• Steel-can ≠ drop-in replacement. The stacked electrode process is a manufacturing change, not a BOM swap. Supply chain qualification at 7mm widths is non-trivial.
• Impedance wins don't scale linearly. At larger cell sizes, the stacked-layer advantage narrows because the jelly roll's impedance penalty shrinks.
• Sustained load is a different beast. A display that draws continuous power (like Ray-Ban Display) needs different cell sizing than a burst-workload device like a camera-only wearable.
• Two-battery systems need symmetric thermal design. If one temple runs hotter, cell aging diverges and the pair goes out of balance over the product's lifetime.

Where This Fits in the Broader Trend

This is part of a larger pattern of hardware teams borrowing techniques from adjacent domains — battery engineering is starting to look more like semiconductor packaging, with stacked layers, tight tolerances, and system-level co-design. For a parallel look at how the software side of large-scale systems is evolving, see Microsoft's 2026 database vision — the same "co-design everything" philosophy shows up there too.

What to Take Away

If you're building any battery-powered wearable, three lessons transfer directly:

1. Pick the cell architecture for the load profile, not the form factor alone. Burst workloads (camera + AI) and sustained workloads (display) want different cell designs.
2. System efficiency beats chemistry every generation. The 2x runtime jump came from firmware and power management, not new cathode chemistry.
3. Multi-cell systems are a firmware problem disguised as a hardware problem. Budget for the state machine, not just the PCB.

Next Steps

• If you're a firmware engineer: study power management IC (PMIC) arbitration and multi-cell state machines. That's the bottleneck skill for the next wave of wearables.
• If you're a hardware engineer: get comfortable with stacked electrode manufacturing tolerances — they're the new frontier for miniaturization.
• If you're a systems architect: read up on cross-charging prevention topologies in parallel battery systems.

For a deeper look at how another hyperscaler handles a similarly complex systems problem — dynamic configuration at scale — check out Airbnb's Sitar sidecar architecture. The pattern of "coordinate many moving parts without a central bottleneck" is universal.

Battery engineering for wearables is where mechanical, electrical, and firmware engineering collide. The teams that win are the ones that treat it as one problem, not three.