Software-Defined Building: Xpanner's $18M Validates the Retrofit Stack

Yesterday's post was about ADLINK putting public earnings behind the cheap edge AI layer. That was the architectural validation: industrial-grade compute modules in the same shape we deploy, growing 28.7% year over year at audited scale.

Today's post is the business-model validation. And it lands the same week.

On May 14, Xpanner — a 2020-founded Physical AI startup that retrofits construction equipment — closed an $18 million Series B bridge round led by Korea Investment Partners, with KB Investment participating. Total funding to date: $38 million. The press release went out the next morning on GlobeNewswire and got picked up by Crunchbase News, The Robot Report, Robotics & Automation News, and TFN by May 15.

The funding number isn't the point. The unit economics are the point. And the named-competitor list — Built Robotics, SafeAI, Caterpillar's autonomous division, Komatsu Smart Construction — is what makes this a public A/B test on the entire shape of a Physical AI business.

What Xpanner sells, in two sentences

The X1 Kit is a retrofit hardware unit and AI processor that bolts onto a contractor's existing pile driver, excavator, or material handler. Once installed, that machine becomes what Xpanner calls Software-Defined Machinery (SDM) — task-specific autonomy that the customer subscribes to, monthly, per license (piling, material handling).

The customer does not buy a new excavator. The customer keeps the excavator they already own, pays once for the hardware kit, and then pays a recurring fee for the software capability they want to switch on.

Read that paragraph twice. That is, almost word for word, the architecture and the pricing of the small-building monitoring deployments this blog has been documenting for two years.

The unit economics

Crunchbase News got the deal details out of Ryan Park, Xpanner's CFO and co-founder. Here is the revenue line:

And here is the part of the deal that should make every services-and-monitoring practice take a hard look at the words on their landing page:

• Gross margin above 80%. Subscription revenue "flows at near-zero marginal costs" once the kit is installed.
• Zero customer churn since US market entry in 2023. Three years, no logos lost.
• 19 of the top 20 US solar farm EPCs are in active transactions or discussions. Named customers include Mortenson, Black & Veatch, and QCells.
• BuiltWorlds calls Xpanner the only "market-proven company in construction Physical AI."

That is what Korea Investment Partners wrote a bridge check on. Not a humanoid demo. Not a foundation model. Eighty percent gross margin, zero churn, and a one-time hardware install that pulls a recurring software subscription behind it.

The four competitors that sold the opposite model

Here is the unusual part of the Xpanner coverage: the press release and the trade-press follow-ups all name the competitors that tried the opposite approach. Most fundraising stories tiptoe around the prior generation. The Xpanner round didn't.

TFN's coverage put it directly: "all competitors pursued traditional 'build new, replace old' approaches that achieved limited commercial scale." The four named:

• Built Robotics — sold autonomous machine pilots. Better-funded than Xpanner in early rounds. Has been pivoting business models since 2024.
• SafeAI — same approach, autonomous heavy equipment as a unit sale.
• Caterpillar autonomous division — fleet replacement strategy, OEM-led.
• Komatsu Smart Construction — OEM-led, mandatory new equipment for the smart-construction tier.

All four sold the customer a new machine — or a software subscription that required them to buy a new machine. Xpanner sold the customer a kit that bolts onto the machine they already own, then sold software on top of that kit.

Same era. Same vertical. Same physics. Different business model. The one that retrofits existing assets and prices per task is at $21M revenue heading to $60M ARR. The four that replace the assets are still working on the model.

Why this matters for a small commercial building

Construction equipment is a different vertical from a community-center boiler room. The customers are different. The actuators are different. The risk model is different. None of that changes the structural lesson here, which is about business-model architecture, not vertical specifics.

Every small commercial building monitoring proposal we write runs into the same buyer question: "why this architecture instead of buying a cloud-connected BMS upgrade?"

Until last week, the answer to that question was:

1. The cloud BMS costs $30,000 to $80,000 per building.
2. It requires a forklift install — rip out the old controls, run new cable, train the staff.
3. It depends on the internet to function — when the connection drops, so does the building intelligence.
4. The lock-in is ten years deep.

Those were the rhetorical answers. Correct, but rhetorical. The buyer would nod, then ask if anyone reputable was building the other model.

Starting this week, the answer is: yes, and they just raised $18 million on it, and they are profitable, and the customer-replacement model lost the A/B test in their vertical.

The same shape, at building scale

Here is what Xpanner's architecture looks like next to ours, side by side:

That table is, line for line, the same business. Different scale, different actuators, different customer vocabulary. Same architecture, same pricing geometry, same reason it wins against the rip-and-replace incumbent.

Software-Defined Building

Xpanner's branding term for the X1 architecture is Software-Defined Machinery. It is a good term. It says, in three words, that the asset you already own becomes the platform and the value sits in the software layer above it.

The parallel term for our practice has been sitting unclaimed on the shelf. Use it now: Software-Defined Building.

The boiler stays. The chiller stays. The roof and the walls and the rooftop unit stay. What changes is the sensor floor, the local compute coordinator, and the software running on top — sized to the building, priced per month per tier. The building becomes a platform that gets smarter through software updates instead of capex cycles.

That is what the agentic sump pump has been doing in a New York basement for two years. It is what the 40-device community-center deployment does at one order of magnitude more sensor count. Both of them are Software-Defined Buildings, retroactively. Both of them ride the same architectural and pricing shape that Xpanner just put a Series B sticker on.

What goes in the next proposal

Until last week, the building-monitoring proposal had to lean on price points and field results. Last week, after ADLINK's Q1 numbers, it could lean on an earnings line. This week it gets a business-model proof point too.

The proposal-language paragraph for this week:

The closest public business-model parallel to this proposal is Xpanner, which retrofits existing construction equipment with edge AI and prices it as a per-task subscription. Xpanner is on track to $60M ARR in 2026 at gross margins above 80% with zero customer churn since 2023, and is in active transactions or discussions with 19 of the top 20 US solar farm EPCs. Four better-funded competitors — Built Robotics, SafeAI, Caterpillar's autonomous division, and Komatsu Smart Construction — pursued the opposite "sell new equipment" model and have not reached comparable scale. The architecture in this proposal is the same shape as Xpanner's, applied to small commercial buildings instead of heavy construction machinery.

That paragraph belongs above the line where the buyer asks why we don't just sell them a cloud BMS upgrade.

The pattern, five weeks running

The cheap-and-retrofit pattern is now visible at five-week resolution.

Five weeks. Five top-of-stack stunt headlines or standards debates. Five sensor- or business-model floor releases that materially changed what we can build and how we can price it. The drumbeat is consistent enough now that any small-building Physical AI practice should be calibrating to it explicitly.

What this isn't

Two clarifications, because the framing is easy to over-read.

Xpanner's revenue is not all software. The X1 Kit is hardware, and the hardware sale is part of the cumulative $31M+ revenue figure. The 80%+ gross margin refers to the subscription layer on top, not the kit itself. Same shape applies to our deployments: the hardware tier ($3K–$15K per building) is cost-recovery; the recurring monitoring subscription is the compounding asset.

One Series B is not a vertical-wide trend. Xpanner is the first publicly fundraising "retrofit + subscription" Physical AI company to clearly outpace the "replace" incumbents at audited revenue scale. Whether the next three rounds in the space follow the same shape will tell us if this is a Xpanner-specific result or a structural one. The base rate from the last decade of B2B SaaS says it is structural. Re-check at the next quarter's funding cycle.

What we are watching next

The next milestone is seven days out: the Robotics Summit & Expo in Boston, May 27–28. NXP at booth #536, a Qualcomm Dragonwing IQ-X workshop on industrial PCs and edge controllers, and AI-track sessions from Brain Corp, Roboto AI, and Agtonomy. The press wire that week typically carries sub-$200 edge-AI module announcements. Whatever lands then will get the same treatment here: how does it land in a basement, what does it change in the per-deployment math, and does it push the hardware tier on our landing page any cheaper.

Related reading: ADLINK Just Put Public Earnings on the Cheap Edge AI Layer | Mind Robotics' unicorn has one customer — your building | Advantech shipped the boiler-room stack | Edge AI just became the cheap option | The agentic sump pump