IoT Integration Platforms Compared: AWS IoT vs Azure IoT vs GCP IoT (2026)

The IoT cloud platform you pick shapes your operational economics for the rest of the product’s life. Switching costs are real, integration patterns are sticky, and the marketing comparisons rarely tell you the things that actually matter at 100k devices. Here is the practical 2026 view.

The honest state of the three

AWS IoT Core is the broad-and-deep option. Mature, full-featured, expensive at scale if you don’t tune it. The default choice for most product teams in 2026.

Azure IoT Hub is the right answer when the customer is already on the Microsoft enterprise stack — Sentinel, Defender, Entra ID, Power BI. The integration story sells itself for those buyers.

Google Cloud IoT Core was retired in 2023. Customers on GCP today either use EMQX Enterprise on GKE, Mainflux, HiveMQ, or migrated to Pub/Sub + custom MQTT broker. Treat “GCP IoT” as a self-hosted decision, not a managed service.

Anyone telling you Google has a managed IoT platform in 2026 is selling you something that does not exist.

What you are actually paying for

A managed IoT platform bundles four services:

1. MQTT broker with TLS + device authentication — keeps your devices online, isolated, and authenticated by per-device cert
2. Device registry — a place to manage device identity, attributes, and lifecycle
3. Rules engine — route messages from MQTT topics to other services (databases, queues, functions)
4. Device shadow / digital twin — last-known-state and desired-state for devices

Plus optional:

• Fleet management (firmware updates, configuration push)
• Greengrass / IoT Edge for edge orchestration
• Defender / Sentinel / Security Center integration

Cost reality at 100k devices

Approximate monthly costs for a steady-state fleet of 100k devices, each sending 1 message per minute (4.3 billion messages/month):

• AWS IoT Core: ~$8,000–$12,000/month at list pricing for messaging + connectivity. Add storage, processing, and egress on top — realistic total $20k–30k/month at this scale.
• Azure IoT Hub: ~$10,000/month for the equivalent S2-tier capacity. Similar storage and processing costs on top. Pricing is more predictable than AWS but ceiling is lower.
• Self-hosted EMQX or HiveMQ on Kubernetes: VM costs around $2,000–$4,000/month for the broker tier; you operate it. Cheaper at this scale, more expensive in operational headcount.

Below 10k devices: managed makes obvious sense. Above 1M devices: self-hosted starts to dominate. The crossover is in the 50k–500k range and depends heavily on your team’s operational capacity.

For deeper cost work see our IoT cloud cost post.

How to pick

Pick AWS IoT Core if

• You are already on AWS for the rest of the stack
• You need the broadest ecosystem of supported devices and SDKs
• You want IoT Greengrass for edge orchestration
• You need Device Defender for fleet-wide security signals

Pick Azure IoT Hub if

• The customer is enterprise Microsoft (Sentinel, Defender, Power BI, Dynamics)
• You need DPS (Device Provisioning Service) for true zero-touch onboarding at scale
• IoT Edge is the chosen edge runtime
• The team has stronger Azure than AWS expertise

Pick self-hosted (EMQX, HiveMQ, Mainflux, ThingsBoard) if

• You are on GCP (since IoT Core is gone)
• Cost matters more than operational simplicity
• You have the SRE team to operate a Kafka-class infrastructure
• You need protocol flexibility AWS / Azure don’t give you (CoAP at scale, custom protocols, AMQP 1.0 with specific features)

Pick a verticalised platform (Particle, Soracom, Memfault, Particle, Balena) if

• You are a startup that wants the device + connectivity + management bundled
• You don’t want to operate the platform layer yourself at all
• The total fleet will stay under 100k devices for the foreseeable future

The gotchas

1. Inbound vs outbound message asymmetry. Most platforms charge per-message or per-message-byte. A device sending 1 message per minute looks cheap until you add the digital twin updates flowing back to it, plus configuration push, plus OTA chunks.

2. Egress fees. Streaming IoT data out of AWS or Azure to a third-party service (Datadog, Snowflake, your own analytics platform) is billed at standard cloud egress rates. At scale this dominates the bill.

3. Per-region replication. Multi-region IoT architectures double the connection cost. If your devices are global but your backend is single-region, your latency and reliability profile is worse than the cost suggests.

4. Vendor lock-in via rules engines. AWS rules engine syntax doesn’t translate to Azure routing. Migrating between platforms means rewriting your message-routing logic. Plan integration boundaries to keep this minimal.

5. Device certificate lifecycle. Provisioning is the easy part; rotating certificates across 100k devices when one of yours leaks is a real engineering project. AWS DPS, Azure DPS, and self-hosted alternatives all handle this differently. Pick deliberately.

What we typically do

For a product targeting global deployment in 2026:

• Default: AWS IoT Core, with custom rules-engine actions wrapped in our own abstraction layer (so we can move if AWS pricing changes badly)
• Microsoft-aligned customer: Azure IoT Hub + DPS + IoT Edge
• Cost-sensitive at >100k devices: EMQX Enterprise on GKE or AKS, with Mainflux or our own thin management layer on top
• Startup, fleet under 50k: Particle, Soracom, or Balena depending on connectivity profile

The platform is a bigger commitment than the protocol. We document the rationale, the cost projection, and the exit ramp in a one-page platform decision memo that goes into the repo alongside the architecture diagram.

If you are mid-way through this decision, we are happy to look at the spec for an hour.