Andreessen Horowitz does not write $50 million checks for software that makes spreadsheets prettier. So when the firm led a Series A of that size into a two-year-old Stockholm startup most people have never heard of, the obvious question is what they saw. The answer is plumbing. Literally.

Endra, a Swedish company building AI that automates the design of mechanical, electrical, and plumbing systems, has raised $50 million in a round led by a16z. Notion Capital and Norrsken VC, both earlier backers, came along for the ride. The new money pushes Endra's total raised to roughly $75 million in its first two years, a velocity that says less about the company's revenue and more about how badly the construction world wants someone to fix this.

Here's the part that sounds boring until you sit with it. Every building you have ever walked into was designed, in part, by engineers placing fire alarms, routing circuits, and sizing ducts by hand. Floor by floor. Room by room. The work is repetitive, slow, and increasingly outsourced because nobody wants to do it. Endra wants to drag that entire $150 billion market into software, and a16z just decided the timing is now.

MEP engineering is the connective tissue of the built environment. It covers the systems that move air, water, and power through a structure, and it is some of the most exacting work in the industry. A mistake means a building that overheats, a sprinkler that fails to trigger, or a code violation that halts a project for months while inspectors and lawyers sort it out.

Most of what an MEP engineer does day to day is not creative. It is rote. Place a fire alarm. Run a circuit to the next room. Size the duct. Check it against the local code. Repeat. The a16z partners who backed the deal described it as entering the same value into a long column of cells one at a time, instead of dragging the formula down. Multiply that by every room in a hospital or a hyperscale data center and you start to understand why the work takes months and why so much of it gets shipped to lower-cost markets.

Endra's pitch is that this is exactly the kind of structured, rules-bound labor that AI is good at right now. The platform ingests standard building model files, integrates with Autodesk Revit, and reconstructs the building in 3D. An engineer sets the rules for a fire-suppression system or a power layout, and Endra generates a clash-free, code-compliant design. The company says it cuts work that used to take weeks down to a few hours. Not a marginal speedup. An order of magnitude.

The deeper insight is that the bottleneck was never the drawings. Software digitized the drawings two decades ago. What stayed manual was the reasoning, the thousand small decisions that turn a blank floor plan into a system that passes inspection. That gap between digitized output and still-human judgment is precisely where Endra has planted itself.

Timing matters here more than usual. The world is building data centers faster than it can design them, and data centers are MEP-heavy buildings. Strip away the racks and the marketing, and a data center is mostly cooling, power distribution, and the redundant systems that keep both from failing at three in the morning. The engineering bottleneck is real, and it is getting worse as hyperscalers race to stand up capacity for AI workloads.

That is the wedge. Endra is not trying to win over every architecture firm on the planet on day one. It is going after the projects where the MEP design burden is heaviest and the cost of delay is most painful. A data center that comes online three months late is a data center that misses an entire planning cycle of demand, and in the current AI build-out, three months is an eternity.

The same logic extends to factories, hospitals, and the wave of reshored manufacturing now underway across Europe and North America. These are the buildings where MEP complexity is the schedule. Compress the engineering, and you compress the whole project. That is a value proposition a CFO understands without needing a demo.

It rhymes with a broader pattern we keep seeing in Nordic deep tech, where startups attack unglamorous infrastructure bottlenecks that suddenly matter because of AI demand. Finland's Quanscient is doing it for physics simulation, and the AI cloud infrastructure land grab is pulling nine-figure rounds across the region. The connective theme is that the picks-and-shovels layer of the AI economy is where a lot of Nordic founders are quietly winning.

Endra was founded by Niklas Lindgren, who serves as CEO, alongside Anton Juric, now president and COO, Gustav Hammarlund as chief product officer, and David Rydberg. The team comes out of Sweden's software scene rather than traditional construction, which is part of the bet. They looked at a 25-year-old design process and decided that being outsiders was the advantage, not the handicap.

Lindgren has been blunt about the core insight. The fundamental way buildings get engineered has barely changed in a quarter century, even as the tools around it went digital. Revit made drawings easier to produce. It did not make the underlying design decisions any faster. An entire profession has been waiting for someone to automate the judgment, not just the documentation.

The capital efficiency story is what likely sealed the a16z deal. Raising $75 million in two years is not, on its own, remarkable in a frothy market. Building a product that large enterprise customers actually deploy in that window is. The company has been deliberately quiet, working with design firms rather than chasing headlines, and that discipline shows up in how cleanly the round came together. When a16z's enterprise team goes public with a thesis, it usually means the diligence calls with customers went better than the founders dared hope.

There is also a talent flywheel forming. The Nordic region has spent a decade producing software engineers who cut their teeth at Spotify, Klarna, and a generation of B2B SaaS companies. Pointing that talent at a physical-world problem like construction is the kind of move that only works when the founding team can credibly recruit people who could otherwise join any AI lab they wanted.

Endra is not the only company circling construction with AI. A clutch of US and European startups are building copilots for architects, generative tools for floor plans, and automated code-checkers. What separates Endra is where it has chosen to fight. MEP is narrower, deeper, and far more punishing to get wrong than aesthetic layout work, which means the barrier to a credible product is higher and the defensibility is stronger.

The moat, if there is one, is the feedback loop. Every project Endra completes teaches the system more about how real engineers solve real problems across real codes. That compounding knowledge is hard for a generic large language model to replicate, because the training signal lives inside the messy specifics of building regulations and field experience rather than on the open web. a16z's enterprise partners have made versions of this bet before, and the pattern they look for is exactly this: a structured domain, a painful manual process, and proprietary data that gets better with use.

There is a reason so many of these structured-domain bets are coming out of the Nordics specifically. The region has a deep bench of engineers comfortable working at the intersection of physics, software, and regulation, the same combination that powers companies like Quanscient. Construction is just the latest domain to get the treatment.

The risk is obvious. Construction is a slow, relationship-driven industry that has eaten plenty of well-funded software startups. Selling automation to engineers whose jobs it partly replaces is delicate, and code compliance varies wildly by jurisdiction, sometimes from one city to the next. Endra has to be right not just often, but every single time, because a hallucinated fire-suppression system is not a typo you patch in the next sprint.

Still, the setup is hard to argue with. A massive, underserved market. A genuine AI use case that is structured rather than speculative. And a demand spike from data centers that nobody serious expects to slow down this decade. a16z is betting that the most boring part of building design is about to become one of the most valuable, and that the team to capture it is sitting in Stockholm.

If Endra is right, the next time you walk into a new building, some of the systems keeping you cool and safe will have been designed in an afternoon by software out of Sweden. You won't notice a thing. That's rather the point.