Memory is profit: choosing a language in embedded is really a cost decision

After many years building routers, one of my deepest lessons:

In hardware products, RAM is BOM, and BOM is profit.

So "Python or C / Rust" was never a technical preference — it's a cost calculation. Yet most discussions stop at "which language is better" and never do the math.

The math: the delta of 128 / 256 / 512 × shipment volume

Using common tiers (numbers are illustrative, don't nitpick the exact price):

Memory tier	DRAM delta per unit (illustrative)
128 → 256MB	about $0.3 – $1
256 → 512MB	about $0.5 – $1.5

Tiny per unit — until you multiply by volume:

100,000 units × $0.5 = $50,000

Meanwhile, "the dev labor Python saves" is a few tens of thousands at most. At volume, the material saved outweighs the labor saved. That's how hardware companies do the math.

Memory isn't just money

For 24/7 devices, memory is also:

• Power / thermals / size (smaller memory → less power → smaller PSU/cooling);
• Reliability (tighter memory → more OOM / jitter — a hazard on shipped devices).

Less memory = less power, less heat, fewer failures.

So "Python's dev efficiency" has to be counted this way

Python saves labor / time; it costs per-unit material:

	Python	C / Go / Rust
Dev labor	✅ less	❌ more
Per-unit memory	❌ costly (interpreter ~tens of MB)	✅ lean
Winner at volume	—	✅ material > labor

The axis is not "which language is better", but:

"volume × per-unit memory delta" vs "dev labor".

What that axis implies

Scenario	Winner	Why
High-volume shipped hardware (routers/consumer)	C / Go / Rust	material > labor
Low-volume / high-complexity / internal tools / cloud	Python	labor > material
Prototypes / glue / tooling	Python	fast, and not in the shipped BOM

So in reality it's always layered: tight-memory devices in C/Go/Rust, Python upstream for orchestration / tooling / non-shipped devices — Python isn't obsolete, it was just standing in the wrong place.

What about Rust? An ideal — not yet in hand

Rust's pitch is exactly "both at once": as safe and easy as a modern language, with a runtime close to C — in theory it eats both "dev efficiency" and "low memory".

But be honest: Rust's dev efficiency hasn't caught up with Python yet (especially prototypes / glue), and there's a learning curve. So "Rust = both" is the direction, not the present. For tight-memory devices, C/Rust are the real answer; where tens of MB are affordable, Python still wins.

Closing

Engineers should not only write code, but do the math:

One device with 128MB more memory = how many DRAM chips = how many dollars → × volume = your profit.

Memory is money. Choosing a language is how you spend it.

💬 How does memory factor into cost in your product? Leave a comment below, or open an Issue to talk.

（Bilingual post.）