The Smart Home Bubble Popped. The Sump Pump Kept Running.

Hackaday published the smart-home obituary on Thursday. The piece is called Why The Smart Home Bubble Popped, and the post-mortem reads about how you would expect: a decade of cloud-dependent platforms shut down, abandoned, spectrum-jammed, or held hostage by quarterly earnings calls. The honest verdict at the end is one sentence: the only thing still working at the end of the decade is the locally hosted alternative, and that alternative is Home Assistant.

The post is not surprising. It is mainstream press finally catching up to what local-first practitioners have been quietly building on for years. What is interesting is that the same week the obituary was published, the sump pump in the basement of a New York house I have been monitoring for two years kept pumping. The 40 sensors in the upstate New York community center kept reporting. Nothing in either deployment depended on a cloud service that could be shut off. Nothing went dark.

That is the entire pitch for the small-building stack in 2026, in one news cycle.

What the obituary actually says

The Hackaday piece does not break new ground. It collects the receipts.

• The first wave of cloud-bound smart-home platforms — Revolv, Iris by Lowe's, Wink (functionally), Insteon (briefly), Best Buy Connect, the original Nest API — went dark or got crippled when their corporate parents decided the business did not pencil.
• The second wave shipped at scale but jammed the 2.4 GHz and sub-GHz bands so hard that consumer-grade Zigbee and Z-Wave deployments started misbehaving in dense neighborhoods.
• The third wave (Matter) is a standard on paper but a hostage situation in practice: every major vendor implements it just enough to keep regulators happy and just incompletely enough to keep their walled garden defensible.

Hackaday's conclusion is that the people still doing useful work in this space have all migrated to locally hosted, open-source, no-cloud-required stacks. The article names Home Assistant specifically. The implied operational rule is simpler: if the system you are buying needs a cloud round-trip to flip a light, you have already bought tomorrow's brick.

The boiler-room version never inflated

Here is the part the consumer-IoT obituary missed.

The small-commercial-building version of the same stack — ESPHome on $4 sensors, Home Assistant on a $200 mini-PC, a Jetson-class reasoning module when reasoning is needed, no cloud round-trip in the loop — never inflated. There was no consumer bubble for the boiler room. There was no marketing budget pushing fragile cloud-tethered light switches into a furnace room. The work was always done by operators who needed the system to keep running when the internet was off, because the consequences of a stuck sump pump or a frozen pipe do not pause for a vendor outage.

That is why the post-mortem reads like vindication, not news. The agentic sump pump in the basement has been running on an ESP32 + Home Assistant + local edge module stack for two years, including through three home-internet outages and a Comcast city-wide failure last winter. The 40-device community center deployment reports through a local coordinator. Neither system has ever called a cloud service to decide whether to run a pump.

The Hackaday post is the first time a high-traffic outlet has named that architecture as the one that survived. Building owners who got burned by a 2021-era smart-home product going dark now have a public-facing article they can read that says: yes, that happened, and yes, the local-first stack is the working replacement.

What the survivor stack looks like in a small commercial building

The pieces are unglamorous and the bill of materials is short.

The architecture is boring on purpose. The only thing that is allowed to be exciting is the part that produces the insight: the vibration signature on the pump motor, the gas-meter pulse rate, the differential pressure across the air handler. The plumbing underneath has to be the kind of thing a maintenance technician can power-cycle without thinking about it.

That is also why this stack survived the bubble. Every layer in the table has at least two viable alternatives at every price point. Nothing in the bill of materials is sole-sourced from a company that can disappear next quarter.

The pattern this week, three sources

If you only watched one news source this week, you saw one story. If you watched three, the story was the same one three times.

The funding number is real. PitchBook reports robotics and physical AI raised $27.6B across 1,009 deals in 2025, more than double 2024. Physical Intelligence is closing a $1B round at $11B valuation. None of that capital is operationally close to your boiler room this month. Some of it will be, in 2028 or 2029. The cheap layer is operational right now.

What this is not

Three clarifications, because the obituary framing can be read too triumphantly.

Home Assistant is open source, not free. Operating it in a commercial deployment costs technician hours. The savings versus a cloud-dependent platform are real, but they show up as predictable maintenance cost rather than as a zero on the invoice. Anyone selling "free smart building" is selling the same fantasy that killed Revolv.

The cloud has not gone away. Remote access, fleet-wide dashboards, and the reasoning layer that compares one building to a hundred buildings are all easier in the cloud. The right architecture is local-first for the control loop and cloud-optional for the analytics and dashboards. The Hackaday obituary is about the platforms that put the control loop in the cloud and then turned the lights off. Do not over-correct.

The bubble pop did not kill all consumer IoT. Plenty of new devices are still shipping with cloud-only firmware, and plenty of building owners will buy them in 2026 without reading the obituary. The opportunity is to be the operator they call when the next platform shuts down.

What we are doing about it this week

Three concrete things on our end, this week, in response to the obituary.

1. The cold-email sequence got a new opener. The first line of the next batch is the Hackaday verdict, followed by the edgeRX number. The pitch closes on a 15-minute call about a single building. We will see in two weeks whether the opener moves reply rate.
2. The sump-pump case study got a new section. Two years of continuous operation, three documented internet outages survived, zero cloud dependencies in the control loop. The post-mortem framing makes the "two years" line land harder than it did last month.
3. The retrofit price list got a "legacy device bridging" line item. Roughly $300 per device, including a $5 ESP32, an ESPHome flash, the serial or Modbus cable, and the integration work to surface the data in the monitoring dashboard. That fee pays for itself the first time the bridged data catches an off-spec reading.

None of this is novel. All of it is now easier to sell.

Related reading: The agentic sump pump | Home Assistant 2026.5 plugged ESPHome into every serial port | Three things stopped building monitoring; fixed-sensor pricing fixed them | Cloud latency is a hazard for robots — and for buildings