What you actually pay for AWS IoT, Memfault, and Balena (2026 edition)
Real IoT platform pricing from vendor filings, HN threads, and engineer-disclosed bills. AWS IoT Core, Memfault, Balena, Particle, and the DIY math.
$34,167 a month. That is what 1,000 industrial sensors cost on AWS IoT Core in EMQX's 2025 cost study, sending 10 messages per second of half-kilobyte payloads. The same workload migrated to a self-managed broker dropped to about $6,000 a month. The 80 percent gap is not a marketing claim. It is what one customer paid before they walked.
That number is not the worst one in this post. The worst one is Memfault's, and the math is public now because Nordic Semiconductor paid $120 million for them in June 2025 and disclosed the financials. $7.2 million of ARR across 100 customers. About $72,000 in average annual contract value, twice their published list price.
This post is for the hardware founder staring at an AWS IoT bill that doubled last quarter, or the CTO whose Memfault renewal just landed and is trying to remember what they actually use it for. Every number below is sourced from public threads, vendor filings, or competitor cost analyses, and linked back.
We sell an IoT platform. We are biased. The numbers below are not.
AWS IoT Core pricing: what teams actually pay at scale
AWS IoT Core is the default. Most teams start there because they already have an AWS account, the docs are good, and the free tier covers the first 50 devices. The pain shows up at scale, in two specific places.
The first is the 5KB billing chunk. AWS bills MQTT messages in 5,000-byte units, rounded up. A 50-byte sensor reading costs the same as a 5,000-byte one. EMQX's cost-analysis post walks through the math: 1,000 industrial sensors at 10 messages per second with 0.5 KB payloads runs $26,280 a month for messaging alone, plus $7,884 for the rules engine, totalling $34,167. Each sensor's actual data fits in a tenth of a chunk. You are paying for the other 90 percent of bytes you never sent.
The second is the per-feature pricing model. Messaging, Rules Engine, Device Shadow, Device Management, Jobs, and Fleet Provisioning are billed separately. Hubble Network's comparison study put 10,000 low-rate devices on AWS at $596 a month for base messaging, plus roughly $1,000 a month once you add Device Management for OTA. Azure IoT Hub came in at $751 a month flat for the same fleet because direct methods and twin updates are bundled.
AWS IoT Core cost by fleet size, at a typical industrial message rate:
| Fleet size | Message rate | Reported $/mo | Source |
|---|---|---|---|
| 50 devices, 12-mo free tier | low | $0 | AWS free tier |
| 1,000 sensors | 10 msg/sec | $34,167 | EMQX 2025 study |
| 10,000 devices | low | $596 base + $1,000 OTA | Hubble Network 2025 |
| 50,000 devices | architected serverless | $750 (messaging only) | Medium walkthrough, 2024 |
| 100,000 devices | scaling | $30,000 to $80,000 | EMQX competitive analysis |
AWS IoT is cheap with a few devices sending little data and expensive the moment your sensors talk often or your payloads are small. The 5KB chunking is the trap most teams do not model in their seed deck.
Memfault pricing: what the Nordic acquisition revealed
Memfault is the device-observability incumbent. They built a real product. The pricing was opaque on purpose until Nordic Semiconductor's June 2025 acquisition disclosure made the math public.
$120 million purchase price. $7.2 million in ARR. 100 customers. That works out to roughly $6,000 per customer per month, or twice the published "Growth" tier of $3,495.
The Vendr marketplace data shows enterprise transactions ranging from $12,000 to $45,000 a year, with the average at $32,000. Both numbers are below the Nordic-disclosed average, which means the long tail of Memfault customers is paying meaningfully more than the marketplace floor suggests.
A pedalpete comment on the HN acquisition thread said they were "struggling with the pricing model for long-term growth". Capterra and SoftwareAdvice reviews flag "pricing is opaque" as a recurring theme.
Memfault is good engineering and the price reflects that. At 100,000 devices with a dedicated firmware-observability budget line, $72k a year is reasonable for what you get. At 1,000 devices wanting OTA plus crash dumps plus a dashboard, the floor is high relative to the use.
Balena, Particle, and Golioth pricing: the per-device tier
The mid-tier platforms are where the surprise-bill stories cluster.
Balena. Published pricing puts 10 devices in the free tier, then $1.50 per device per month after a 2023 pricing update that the community reaction post tried to soften. Mid-tier customers absorbed the increase without much warning. The "let us explain" tone of the post is a tell.
Particle. $2.99 to $5 per device per month on the Basic plan, plus cellular overages if you use Particle SIMs. The community forum has a recurring complaint that the cloud-services fee feels high if you bring your own SIM.
The Particle story that made the rounds is from a February 2024 community thread. A user named risingtiger ran cellular devices configured to send 4 messages a day, each under 1 KB. They burned 10 to 21 MB per month each because they were in poor signal locations and reconnecting constantly. Particle's account-deactivation threshold triggered at 360 MB cumulative. The account was shut off. Particle staff response: "Devices that are really struggling with signal are likely disconnecting and connecting endlessly and consuming data overhead." Translation: working as designed.
Golioth simplified their pricing in April 2026. $50 of free credits then $0.25 per device per month plus $0.0095 per MB of OTA. Cleaner than Particle. Still per-byte billing under the hood, which means the same misbehaving-device risk applies.
The pattern across all three: per-device list prices look benign, the actual bill depends on a usage axis you do not fully control. Signal quality, payload sizes, OTA frequency, reconnect storms. None of that shows up in the marketing-page calculator.
The hidden cost: surprise bills
This is the category that does the real damage to a startup, and it is not unique to IoT.
A Vercel customer was hit with a $23,000 bill in 2024 after a DDoS attack pushed 5 TB of malicious traffic through their site. Vercel charges all served bandwidth at $0.15 per GB, including DDoS traffic. The HN comment that lit up the thread: "this is my worst nightmare as a bootstrapped founder." Vercel's Hobby tier caps cost by pausing the site instead of billing. Pro just keeps the meter running.
The IoT version is the AWS S3 egress bill that ate someone's seed funding because their telemetry pipeline accidentally re-uploaded historical data on every gateway restart. We will not link to a specific story because the founders involved generally do not want it indexed. Ask any solo founder who shipped on AWS IoT in 2023 and you will hear at least one.
The shape of the problem is the same in every case. Your cost is multiplied by a variable you do not control: malicious traffic, reconnect storms, an upstream vendor's behavior. Per-device-per-month billing is the antidote. A misbehaving device costs you a flat $25 of headroom for the month. Not $25,000.
DIY IoT platform cost: the math nobody puts in the deck
The other path most founders consider is "we will just build it ourselves." The math on that is also worth being honest about.
A senior embedded engineer in the US in 2026 costs $140,000 to $170,000 base, per Glassdoor's May 2026 data. Loaded cost (benefits, payroll tax, equipment, recruiting fees) runs 1.3x to 1.5x base. The all-in first-year number is $185,000 to $230,000. Add a 15 to 25 percent contingency-recruiter fee on top and you are at $220,000 to $270,000 for the first year of one engineer. We unpacked this in detail in the real cost of an embedded engineer.
You usually need 1.5 of them, because the work spans firmware, cloud transport, OTA, observability, and security, and one person cannot ship all of it without burning out. So call the loaded annual cost $300,000 to $400,000 for the team.
Then add the platform fees they are running on. Self-hosted ThingsBoard CE on a Hetzner box costs €40 to €80 a month flat (a real community-reported number) but assumes 0.5 FTE of operator time to maintain it. Managed services cost more in cash, less in headcount.
Then add the time. Six to eighteen months of glue work between the broker, the cert pipeline, the OTA system, the observability stack, the dashboard, and the customer-system integrations. We wrote about the connected-product penalty at length: every month the cloud side lags the hardware is a month of revenue compression and deal-size shrinkage.
The line item that ends up in the board memo is "embedded engineer: $200,000." The line item that should be there is "device-to-cloud capability, all-in: $400,000 to $600,000 in year one, plus 9 months of opportunity cost." The salary aggregators do not show that number because they only know about salaries.
What we charge
Here is the SCADABLE number, since you asked. It is not the point of this post and it is on the pricing page in more detail.
We charge per device per month, on a logarithmic curve indexed by data rate. $8 for a sensor that reports once a day. $25 for one that reports once a minute. $60 for a high-rate stream at 10 messages per second. The curve compresses on purpose because the customer's value perception scales with operational risk, not message count.
The platform base fee is $0 for personal accounts (5 devices, no AI agent), $199 a month for Starter, $1,499 a month for Business with the AI onboarding agent unlimited. Enterprise is sales-led and sits in the $60,000 to $120,000-a-year range, which is below Memfault's average ACV by design.
The AI onboarding agent (the thing that turns a device datasheet into a working integration) is paid-only. We are not training the market to expect it free. Linear, Notion, Vercel v0, and GitHub Copilot all gate AI features behind paid plans for the same reason: real LLM tokens cost real money, and the customers who get value out of it are the ones who would pay anyway.
The two commitments we make in writing on the pricing page:
- No surprise bills. Per-device-per-month is the cap. A misbehaving device in a poor-signal location costs you the per-month rate, not unbounded data charges. We will warn at 60, 80, and 100 percent of any quota and grace you for 7 days before any cutoff. We will never silently deactivate the way Particle did to risingtiger.
- Founding-customer pricing for the first 10 customers. Flat $500 to $2,000 a month for unlimited devices up to 1,000, in exchange for a logo and a case study. Verdant's deal is in this category. We will honor it. We will not raise on you in year two.
How IoT platform pricing compares in 2026: send us your bill
If you are running a connected hardware product in 2026, you are probably paying one of the bills above. We want to publish anonymized comparisons. Send us a screenshot of your last AWS IoT bill, your Memfault invoice, your Balena overage line, your Particle deactivation email. We will redact the customer name, normalize the units, and add it to a public dataset other founders can reference when they are doing the same math you are doing now.
The market has been opaque about IoT-platform pricing for too long. The reason your AWS bill doubled is not that AWS got greedy. The reason is that nobody told you the 5KB chunking trap before you architected around it. The reason Memfault feels expensive is that the published price is half the average. The reason your Particle account got deactivated is that the billing unit (cellular bytes) is not the unit you were monitoring (application messages).
The fix is shared data. Send us yours.
If you want help modeling what your specific fleet costs across these platforms, or what it would cost to move it, we run 30-min architecture reviews for hardware founders making exactly that call. Book at https://cal.com/rahbaral/quick-chat