12/06/2026
In 2023, Google gave its Cloud IoT Core customers one year's notice – then switched it off.
Everyone who'd built on it had to re-platform their entire fleet.
That's the part of connected hardware nobody budgets for. The plumbing you rent can vanish, and the data architecture you build freezes far faster than you expect.
Here's the trap. Early on, streaming raw telemetry to the cloud is almost free — pennies at ten devices. So teams ship everything upstream and move on. That bill is invisible at ten units and a recurring tax at ten thousand. By the time it bites, the architecture has hardened around the cheap early assumption, and clawing processing back to the edge means new hardware in the field — which you can't push over the air.
Two questions decide this, and most teams answer them by accident: what do you measure, and where do you run it? They look like separate calls — one for firmware, one for the cloud team. They're the same decision in different currencies. More logic on the device means costlier hardware. More raw telemetry means a cloud bill that compounds with every unit you ship.
The discipline that survives fleet scale is easy to say and rare to do: rent the commodity, own the differentiator. Connectivity, storage, dashboards — buy them, every rival can reach the same tools. The data your fleet generates and the models trained on it — own those, or you've left your moat in someone else's cloud.
So size it before you sign. Real numbers on device count, payload, and cadence — before the vendor contract, not after the bill.
Where have you seen this freeze hardest — the cloud bill, or the hardware you couldn't change once it shipped?
For a connected product, what to measure and where to process it are the same decision in different currencies. A field guide to the five data classes, the…
