eSIM for IoT Fleets: Removable eUICC vs Soldered MFF2

Short answer: pick your form factor from the failure mode you expect, not the unit price. The hardware delta between removable industrial eUICC and soldered MFF2 is roughly $2–4 per device. A single physical SIM swap in the field costs $50–500 per device. That 20-to-100x asymmetry means the real question is: if a profile, carrier, or SIM fails, can you reach the device? MFF2 if the device is sealed, mobile, unattended, or theft-exposed. Removable industrial eUICC if it's serviceable, low-volume, or a retrofit. And separately from form factor: if your devices are headless, you need SGP.32 and an eIM, because the consumer eSIM provisioning flow does not work without a user.

Disclosure: Switch eSIM sells removable eUICC hardware. This article recommends soldered MFF2 for several common fleet profiles, and recommends against our consumer-grade product line for fleet use entirely.

First: three terms that get conflated, and one that will cost you money

MFF2 is a package, not a capability. It's a 6 × 5 × 0.82 mm solder-down chip in a DFN8/QFN8 package, specified in ETSI TS 102 671. That's all it says.

eUICC is the software capability — the ability to hold and remotely switch operator profiles. Keepgo states it directly: "An MFF2 SIM can support eUICC, but it doesn't have to." Onomondo is blunter: "SIM hardware and SIM software are independent choices, and understanding that distinction is critical for IoT deployments."

Four combinations exist, and all four ship commercially:

	Removable (2FF/3FF/4FF)	Soldered (MFF2)
UICC (fixed profile)	Standard IoT SIM	Embedded fixed-profile SIM
eUICC (remote provisioning)	Industrial removable eUICC	Industrial MFF2 eUICC

The distinction that actually costs money: commercial grade vs industrial grade. Both exist in both form factors, and the spread is significant:

	Commercial grade	Industrial grade
Operating temp	−25°C to +85°C	−40°C to +105°C
Typical use	Wearables, indoor sensors	Automotive, metering, outdoor, agriculture

Soracom's guidance: commercial grade suits "smart devices, wearables, general-purpose sensors… where the temperature, humidity, or other environmental conditions are not so severe that a regular SIM card would fail prematurely." Industrial grade is for "agriculture, smart factories, certain types of transportation."

Consumer-grade eSIM adapter cards — the kind sold for phones, including ours — are not on this table. They are not rated to industrial temperature, and more fundamentally they are managed by a phone app requiring OMAPI. A headless tracker has no phone and no app. Do not deploy consumer adapter cards in a fleet. They will fail on temperature, on vibration, or on the simple fact that nothing can manage them.

What SGP.32 changes, and why most IoT eSIM content is out of date

If your devices are headless, form factor is the second decision. The first is which provisioning standard your eUICC and module support.

	SGP.02 (M2M)	SGP.22 (Consumer)	SGP.32 (IoT)
Designed for	Static industrial M2M	Phones, tablets	Headless IoT fleets
Activation	Server-to-server push	User scans QR	Server-orchestrated, no UI
Transport	SMS + HTTPS	HTTPS/TCP	TCP/IP or CoAP over UDP with DTLS
Key limitation	Structural operator lock-in	Requires a human	Interoperability still maturing

Why SGP.02 became a trap. The lock-in was architectural, not commercial sloppiness. Per 1ot's account: every M2M eUICC is bound to a single SM-SR, and that SM-SR can only deliver profiles from SM-DPs it has been pre-integrated with — integrations "typically negotiated between competing operators, who have limited incentive to make it easy." emnify's assessment is that for most deployments, "large-scale remote profile swapping simply wasn't commercially feasible."

What SGP.32 introduces:

eIM (eSIM IoT Remote Manager) — server-side orchestrator that triggers download, enable, disable, delete across a fleet via API
IPA (IoT Profile Assistant) — on-device component that executes profile operations
SM-DP+ reused from the consumer spec, collapsing SGP.02's SM-DP and SM-SR into one

The practical wins for constrained fleets: no SMS dependency (BICS), and CoAP over UDP with DTLS, which matters concretely on NB-IoT and LTE-M where SGP.02's SMS-and-HTTPS assumption breaks battery and bandwidth budgets.

Two honest caveats. First, vendor descriptions of the mechanism conflict — BICS calls it "a server-driven push approach," Onomondo says it "takes the remote provisioning from SGP.02 and the pull model from SGP.22." Both are loosely describing a system where the eIM initiates but the IPA performs the download. Read the spec, not the marketing. Second, on maturity: MVNO-index cautions that "real-world interoperability is still evolving. Differences in eIM implementations, IPA behavior, and module firmware mean that not all solutions are equally mature."

Timeline (single-sourced to 1ot; verify against GSMA before publication): SGP.31 architecture published 19 April 2022; SGP.32 technical specification 29 May 2023; both at v1.2; SGP.33 test specification published in three parts on 27 January 2025 — SGP.33-1 (eUICC), -2 (IPA), -3 (eIM and SM-DP+).

Field-swappability vs tamper resistance

This is the real trade, and it's genuinely two-sided.

The case for soldered: theft is a live operational problem, not a theoretical one

SIM theft from fleet devices is common enough to have its own remediation literature. Freeeway describes the pattern: SIMs "removed from trackers and inserted into smartphones or other data-hungry devices," cloned, or "sold on black markets, especially in regions with expensive data plans." The tell is usually financial — "many fleet managers only notice something's wrong when a sudden spike in data usage appears on the invoice."

emnify's recommended first control is the form factor itself: embedded SIMs "are soldered on the device and will break when removed."

If your devices sit unattended in public — scooters, trackers, meters, roadside cabinets, vending — a removable SIM is an attack surface with a direct billing consequence.

Mitigations if you still need removable: IMEI-SIM binding so a lifted SIM won't authenticate elsewhere; concealed slots or epoxy seals; cellular firewalls and per-SIM usage caps. These work, but they're layers you're adding to compensate for the form factor. Note that epoxy-sealing a socket forfeits the field-swappability you chose removable for.

The case for soldered: mechanical reliability

Vendors consistently claim sockets degrade under vibration, dust and thermal cycling — Keepgo: "those points can degrade, especially in environments with vibration, dust, or temperature swings… no connector to loosen or wear out." emnify cites "vibration and shock resistance which is especially relevant for mobile IoT use cases."

Flagging honestly: every source making this claim sells MFF2. The physics is plausible and the consensus is broad, but I found no independent quantified failure-rate comparison. Treat it as well-supported industry consensus, not a measured number. If you're deciding a 10,000-unit deployment on this, ask your SIM vendor for actual field data and get it in writing.

The case for removable: recovery, retrofit, and no SMT dependency

Recovery. A soldered chip that ends up in a bad state — corrupted profile, exhausted profile slots, an eUICC that won't accept a new profile — is a board-level rework job. Replacement means cleaning pads, re-printing solder, and a hot-air rework cycle. If the device is sealed or potted, it may not be economically recoverable at all.

Manufacturing overhead you may not have priced. MFF2 chips are moisture-sensitive SMDs governed by IPC/JEDEC J-STD-020 (classification) and J-STD-033 (handling). 1NCE's MFF2 packaging documentation includes moisture barrier bags, desiccant, and humidity indicator cards for exactly this reason. The failure mode is real: absorbed moisture flashes to vapour above 200°C and can crack or delaminate the package — sometimes with an audible pop. Practically, this means dry-pack storage, tracked floor life, and bake-before-reflow procedures. If you don't already run a moisture-controlled SMT line, that's a new process, not a line item.

Retrofit. Adding cellular to existing hardware with a socket, or swapping a fleet's connectivity without a PCB respin, is where removable wins outright. There is no soldered option for hardware that already exists.

Low volume and prototyping. At 10 or 100 units, reel packaging, dry-pack handling and pick-and-place setup are pure overhead.

Per-device cost: why the unit price is the wrong variable

The finding that should drive your decision. Public unit pricing for eUICC hardware clusters in a narrow band:

Item	Public price	Source
Hologram Hyper eUICC, triple-cut (2FF/3FF/4FF), industrial	$3.00	Hologram store
Soracom industrial MFF2, 100-unit reel	$600 → $6.00/unit	Soracom store
Generic IoT SIM (UICC)	$1–3 by volume	Simplex Wireless
Generic eUICC	$2–4 by volume	Simplex Wireless
Profile download fee	$1–2 per download	Simplex Wireless

And the number that dwarfs all of them:

"A truck roll to go and swap a SIM card physically can run anywhere between $50 to $500 per SIM card." — Simplex Wireless

Run that at fleet scale. Hardware delta between removable industrial eUICC and MFF2 is on the order of $2–4/device — roughly $30,000 across 10,000 units. A single fleet-wide physical swap event at $50–500/device is $500,000 to $5,000,000. Simplex's own comparison: swapping 10,000 physical SIMs costs $500k–$5M, versus $10k–$20k in download fees to do it over the air.

So the decision isn't "which chip is cheaper." It's "how many times will I need to touch this device, and what does touching it cost?" Once you frame it that way:

Devices you can reach cheaply (bench-accessible, indoor, staffed sites) → removable is fine, and the recovery path is worth real money
Devices that are expensive to reach (vehicles, remote assets, sealed enclosures, customer premises) → the form factor barely matters; what matters is that the eUICC and module support remote provisioning at all, so you never make that trip

Cost model to populate with your own numbers. I'm giving you the line items rather than invented figures, because above ~1,000 units every vendor I checked moves to quote-based pricing:

Line item	Removable industrial eUICC	Soldered MFF2
eUICC unit cost	~$3 at list, quote-based at volume	~$6 at 100, quote-based at reel volume
Socket / connector BOM	Yes — measure in your BOM	None
PCB area	Larger	Smaller
Assembly	Insertion (may be manual)	Pick-and-place, one reflow pass
Dry-pack / MSL handling	Not applicable	Required — J-STD-033
Scrap & rework	Card replaceable	Board-level rework
Recovery from bad profile state	Swap card	Rework or write off unit
Theft exposure	Requires mitigations	Structurally low
Profile download fees	$1–2, same either way	$1–2, same either way

Ask vendors for the quote-based cells at your actual volumes and publish the real numbers once you have them. Don't let me guess.

Capacity and lifetime specs worth checking before you commit

Profile capacity is memory-bound, and memory varies more than you'd expect between products — including in counterintuitive directions. Hologram's published figures: SGP.02 Hyper SIM has 414,384 bytes available; the MFF2 SGP.32 variant has 318,936 bytes; the Hyper eUICC card lists 800 KB NVM. iot.cards advertises "up to 10" simultaneous profiles depending on variant.

Check that your profile-rotation strategy fits in NVM on the specific SKU you're buying, not on the vendor's flagship.

Lifetime specs also differ: 1NCE publishes 10 years data retention and 500,000 read/write cycles; emnify and Soracom cite 15 years (emnify specifies 15 years @ 85°C). Different test conditions, so don't compare the headline numbers directly — get the conditions.

Also budget for eUICC's ongoing costs. Onomondo's list of what catches people out: "ongoing costs for inactive profiles, restrictions on profile removals, and mandatory bootstrap profiles. Some providers may charge for the storage or potential activation of these profiles."

And a contrarian option worth pricing. Onomondo argues multi-IMSI UICC beats eUICC for some fleets — eUICC's "advantage comes with higher complexity, power and data overheads, and long-term lock-ins." If your requirement is "work on several networks in one region" rather than "switch carriers commercially over a 10-year life," multi-IMSI may be cheaper and simpler. Price both before assuming eUICC.

When soldered wins outright

Choose MFF2 without much deliberation when any of these apply:

Sealed or potted enclosure — no service access by design (IP67+, encapsulated meters)
Unattended in public — theft exposure with direct billing consequences
High vibration or wide thermal cycling — vehicle, rail, industrial machinery
Truck-roll cost above ~$100 — the hardware delta is noise
Volume above ~10,000 with an existing SMT line — reel handling is already in your process
Space or BOM-constrained — 6 × 5 mm and no connector
Long design life with no planned service visits — 10–15 year meters and infrastructure

Choose removable industrial eUICC when:

Retrofit into existing hardware — no soldered option exists
Under ~1,000 units — SMT overhead dominates
Prototyping or field trials — you'll want to swap and diagnose
Serviceable devices — a technician visits anyway
You need a recovery path — a card can be replaced; a soldered chip needs rework
No moisture-controlled SMT capability — J-STD-033 handling is a real process cost
Regulatory or customer requirement for physically removable credentials — occurs in some regulated sectors

Our management platform and volume tiers

Section held pending internal specification. For this to be publishable to enterprise buyers I need, in writing:

Is there a GSMA SAS-SM certified eIM, our own or a named partner's? Which?
Which specs do our eUICC SKUs implement, with versions — SGP.02 v4.x? SGP.22? SGP.32 v1.2?
Are the eUICCs SGP.33-certified, and against which parts?
Which grade — commercial (−25/+85) or industrial (−40/+105)? Per SKU.
NVM per SKU and maximum concurrent profiles
What the $814 SKU contains: quantity, form factor, grade, spec support
API surface: bulk provisioning, profile enable/disable/delete, per-device status, webhooks
Which cellular modules we've tested IPA behaviour against

I won't write claims about certification or remote-management capability without these. Enterprise buyers ask for certificates, and a claim we can't document does more damage in B2B than a missing section.

Limitations, stated plainly

SGP.32 interoperability is still maturing. eIM, IPA and module firmware differences mean solutions vary. Pilot before committing a fleet.
eUICC does not eliminate commercial lock-in. Soracom is careful about this: the eIM "doesn't eliminate vendor lock-in entirely (commercial agreements and ecosystem support remain factors)." You still need a contract with the carrier whose profile you want.
Running SGP.02 and SGP.32 in parallel is real overhead during migration, per Idemia's own positioning — relevant if you have deployed hardware on the old spec.
eIM APIs are a new attack surface. MVNO-index: enterprises "must secure eIM APIs, enforce role-based access controls, and monitor provisioning activity to prevent misuse or compromise at scale."
The socket-reliability advantage of MFF2 is vendor-asserted, not independently quantified here.
Volume pricing is opaque above ~1,000 units. Every figure in this article is list price; assume real quotes differ materially.
Consumer-grade adapter cards are unsuitable for fleets — temperature rating, and no host to run the management app.

Testing status

Nothing fleet-specific published yet. Planned: NVM and concurrent-profile capacity measured per SKU rather than quoted; IPA behaviour across common LTE-M/NB-IoT modules; profile download success rate and data cost over CoAP/DTLS on a constrained bearer; recovery procedure for an eUICC in a bad profile state.

Sources: GSMA SGP.02 / SGP.22 / SGP.31 / SGP.32 / SGP.33 specifications · ETSI TS 102 671 (MFF2 package) · IPC/JEDEC J-STD-020 and J-STD-033 (moisture sensitivity and handling) · Hologram IoT SIM datasheets and store pricing · Soracom store and industrial-vs-commercial grade guidance · 1NCE Industrial IoT eSIM datasheet · emnify form factor specifications and SIM-misuse guidance · Onomondo on eSIM vs eUICC and specification lineage · BICS, 1ot, MVNO-index and Soracom on SGP.32 architecture · Simplex Wireless SIM vs eSIM cost analysis · Freeeway on fleet SIM theft · Keepgo MFF2 guide

Learn more:

Back to blog

Item added to your cart