A week after NXT BLD, a16z led Endra's $50M Series A. The post is dated 1 June 2026 and runs to less than 500 words. The number is the headline; the underlying signal is bigger.
MEP was the quietest of the categories I expected to write about coming out of this year's event. It is now the loudest.
The Three at NXT BLD
The three companies are coming at MEP from different ends of the market, and the contrast between them is more useful than any one of them is in isolation.
Augmenta is the most ambitious. They frame what they are building as a foundation model for construction, with the underlying line that building design is a geometry problem, not a language problem. They have started with electrical, and they are running real work at datacenter scale, which is the hardest electrical coordination environment in the industry. Mechanical and plumbing are flagged as coming. If they execute on the breadth they are signalling, Augmenta sets the upper bound of what AI in MEP can be.
HVAKR is at the other end. Cloud-based HVAC design — load calculations, takeoffs, ASHRAE templates, reports — for the firms still doing this work in spreadsheets and on-device legacy tools. Narrow scope, accessible pricing, and a customer list that suggests the mid-market has already started to move.
Endra sits between them. Broader than HVAKR, narrower than Augmenta's foundation-model framing. Revit-native, fire alarm shipped first, the rest of the MEP scope in the pipeline. A customer list that reads as a Who's Who of European MEP consultancy. And the company a16z just put $50M behind, which sharpens everything else about their position.
What These Tools Actually Compress
It is tempting to make the same argument here that I made for Branch — that the value sits in bringing the engineering conversation upstream into early-stage architectural iteration. It would also be wrong.
MEP is not a phase you can usefully run early. The project needs direction first. Programme, massing, structural grid, basic systems strategy — most of these have to be roughed in before placing a fire alarm or running a circuit means anything. MEP belongs after the architecture has committed to enough decisions to be worth coordinating against.
The value Augmenta, HVAKR, and Endra are producing is not earlier MEP. It is the same MEP, much faster. The customer line in the a16z piece is the sharpest framing of it I have seen: basically you could ride a horse with the legacy solution or go to the moon with Endra. The thing being compressed is the rote engineering work — placing thousands of devices, running circuits, sizing ducts, checking against code — that used to consume months of analyst time and is now possible in a sitting.
That speed has knock-on effects that matter more than they sound at first.
Coordination cycles tighten. Clashes that used to surface at week eight of the MEP phase now surface in the first round, when there is still budget and design flexibility to absorb them. Delays stop compounding. The architect does not lose two weeks waiting for an MEP markup that was always going to demand the same five corridor and ceiling-zone adjustments. The MEP consultant does not have to choose between depth of analysis and meeting the issue date.
None of that is glamorous. All of it is what actually slows projects down today, and all of it is what Augmenta, HVAKR, and Endra are now in a position to fix. Different angles on the same underlying question: why is this still so slow?
Why a16z Are Now in the Trade
The a16z piece is short and direct. Their substantive claim is that most of what MEP engineers actually do is rote — place a fire alarm here, run a circuit there, check it against code, repeat — and that this is precisely the work AI is now in a position to absorb. That is the same observation that sits underneath my moat map: MEP design is one of the categories where AI compresses the gap fastest, and where the existing software stack is doing less of the work than the professional accountability layer the engineer's signature represents.
The investment thesis is also the same one a16z published in March and which KP Reddy responded to and which I wrote up in The Missing Player. That piece was the macro case. Endra is the first portfolio bet that follows from it publicly. It will not be the last.
There is also a clarification worth making. The $150B+ a16z attaches to the MEP market is services revenue, not software. It is the money currently being paid to MEP consultancies for the work the platforms above are now compressing. That distinction matters. Endra, Augmenta, and HVAKR are not competing for a slice of a $150B software market — they are competing to absorb a fraction of a $150B services market into software economics. That is the version of the thesis that holds up to scrutiny.
Inside the 20%
This MEP cohort sits inside the 20% of Martin Fischer's 80/20 reframe — the productivity-improvement bucket rather than the workflow-reinvention bucket. Endra, Augmenta, and HVAKR are making the existing MEP delivery process much faster. They are not yet reinventing what the MEP scope is, what an MEP consultant's relationship to a project looks like, or what the owner's role in defining performance becomes.
That is fine. The 20% bucket needs to exist; the 80% one cannot be built without it. But the question I would put to the founders of all three companies is the same question Matt was asking. Once your tool produces a fully coordinated MEP model in two days instead of three months — what does the project's overall delivery model become? Who pays for the lifecycle performance? What does the owner's brief now ask for that it could not before? Those are the harder questions, and the more durable ones.
The headline today is the $50M. The story underneath is that MEP — the category most architects and most software founders have spent twenty years politely declining to focus on — has finally become the most active part of the AEC software conversation.
HVAC is heating up. The pun is intentional.