Vikas Enti spent seven years building the robotic systems that move packages through Amazon fulfillment centers, helping scale the fleet past 500,000 robots across hundreds of facilities worldwide. He watched that infrastructure grow from prototype to planet-scale logistics, one warehouse at a time, each facility tuned to its local demand curve and labor market. In 2022 he left Amazon, took two colleagues with him, and started Reframe Systems with an idea that sounds obvious when you hear it and radical when you see the numbers: build houses the way Amazon moves boxes.
Not in one colossal factory, but in dozens of small ones scattered near the neighborhoods that actually need the houses.
That distinction matters more than anything else in this article, because the graveyard of companies that tried the first approach is deep enough to bury $3 billion in venture capital and still have room for more.
The Graveyard
Start with Katerra. SoftBank's Vision Fund poured over $2 billion into a company that at peak employed 8,500 people and planned 14 distribution centers. Katerra built exactly two manufacturing locations before filing for bankruptcy in June 2021, a timeline so compressed that most of the planned distribution network never made it past the architectural rendering stage. McKinsey's 2019 construction technology report named it a "leading example" of factory-built housing innovation, and within two years the factory had been sold at auction for $21.25 million to Volumetric Building Companies. That sale price, 1% of the capital invested, tells you everything about what went wrong. One percent.
Then there is Veev, Israeli co-founders, $647 million raised, faltered in 2024. HousingWire's post-mortem identified the core paradox: "Solving residential construction's challenges are 5-10 year projects. They don't square up with VC investment cycles." Veev tried simultaneously to discover brand-new fabrication systems, source novel materials, scale distribution across different geographies and climates and building codes, and run profitably. Each of those problems alone would consume a startup, and combined they consumed $647 million.
Skender Manufacturing closed under similar circumstances, smaller numbers but the same fundamental failure mode. Factory-built housing has been under 5% of US new construction for decades, and despite billions invested the needle has barely moved. So when another batch of startups shows up claiming AI will fix it, skepticism is not just warranted but professionally required.
Microfactories: What Actually Changed
Reframe's $20 million Series A closed in September 2025, co-led by Eclipse and VoLo Earth Ventures, with MassMutual, Cubit Capital, RA Capital, Saga Ventures, and Nor'easter Ventures participating. Amy Villeneuve, former COO of Amazon Robotics and someone who personally oversaw the scaling of warehouse automation from hundreds of robots to hundreds of thousands, sits on the advisory board. So does Charly Mwangi, a manufacturing veteran whose resume runs through Nissan, Tesla, and Rivian, the kind of person who has watched factory floors succeed and fail enough times to know the difference between a viable production system and a fundraising deck dressed up as one.
Here is where it gets structurally different from the companies that failed, and I want to be precise about the structural difference because "AI" alone explains nothing.
Katerra's model required massive upfront capital expenditure before a single home shipped: build a megafactory, staff it with thousands of workers, then amortize the cost over years of production at volumes that never materialized because the demand wasn't there yet, because the product wasn't proven yet, because the codes weren't aligned yet, and the whole thing collapsed under the weight of its own fixed costs before it could iterate its way to product-market fit.
Reframe's microfactories stand up in under 100 days. Each one contains specialized "work cells" where teams and robotic systems handle specific production stages. Each factory produces roughly five single-family homes per week, or 30 to 50 per year according to reporting from the Boston Globe. Workers follow iPad-based step-by-step instructions that Enti describes as "like Lego instructions," and the comparison is more precise than it sounds: each step is sequenced, visual, and tolerant of varying skill levels, which means you don't need to recruit experienced framers in a labor market where experienced framers are already spoken for.
And then there is the AI component, which is genuinely novel rather than the bolted-on marketing afterthought that construction tech companies have been slapping onto press releases since ChatGPT made "AI-powered" the most overused adjective in the English language. Reframe's software ingests local zoning ordinances, safety codes, and design requirements, then customizes each home's design to comply automatically. This solves the problem that killed Veev: you don't need to discover a universal fabrication system that works everywhere, because the software adapts each factory's output to wherever that factory happens to be. A microfactory in Somerville, Massachusetts produces homes that meet Massachusetts building codes, while a microfactory in Altadena, California, where Reframe is currently rebuilding homes destroyed in the 2025 wildfires, produces homes that meet California's far more stringent energy and seismic requirements without any manual reconfiguration of the production line. Same system. Different output.
Who Else Is Shipping Product
Azure Printed Homes is the most interesting comparison because it takes the same small-factory thesis and applies it to 3D-printed construction. Revenue hit $7 million in 2025 with positive EBITDA, a detail worth pausing on because Katerra burned through $2 billion without ever reaching profitability. Azure has $62 million in pre-orders, a new Denver factory that opened April 14, 2026, and a partnership with Advantage Homes, California's largest modular housing dealer with eight offices and 150 salespeople. Their first completed development, Welcome Home Village in San Luis Obispo, delivers 62 transitional housing units with a grand opening scheduled for June 8, 2026. Wells Fargo announced mortgage incentives specifically for 3D-printed homes, which is the kind of institutional validation that didn't exist when Katerra was running.
BOXABL is going public through a SPAC merger with FG Merger II Corp. on NASDAQ. Foldable, factory-built homes that ship flat and unfold on-site. Plant Prefab operates a $40 million, 270,000 square-foot smart factory in Arvin, California, the only US facility purpose-built for both panelized and modular construction, with 3 million square feet of annual capacity and automated framing lines. FactoryOS in Vallejo, California, co-founded by Larry Pace and Rick Holliday, partnered with the Northern California Regional Council of Carpenters to produce unionized factory-built multifamily units that cost 20% less and arrive 40% faster than conventional construction, per Harvard Business Review.
The Capital Efficiency Calculation Nobody Published
I ran the numbers because nobody else had. The gap is striking enough to explain why this generation might survive where the last one didn't.
Katerra raised $2.2 billion and built two manufacturing facilities. At peak capacity assumptions, those two facilities were designed to produce perhaps 2,000 to 4,000 units per year combined, though actual production never reached those figures. That works out to roughly $550,000 to $1.1 million in capital raised per unit of annual production capacity, for capacity that was never fully utilized.
Reframe has raised $20 million, and a single microfactory producing 50 homes per year costs a fraction of that to stand up. Even if we assume the entire $20 million goes to a single factory, which it doesn't because Reframe is operating in multiple locations, that is $400,000 per unit of annual capacity. If two factories are operational, the figure drops to $200,000 per unit. At three factories the figure drops to roughly $133,000 per unit of annual capacity.
| Company | Capital Raised | Factories | Est. Annual Capacity | Capital per Unit of Capacity |
|---|---|---|---|---|
| Katerra | $2.2B | 2 | ~2,000–4,000 | $550K–$1.1M |
| Reframe Systems | $20M | 2–3 | ~100–150 | $133K–$200K |
| Azure Printed | Undisclosed | 2+ | ~100+ (est.) | Revenue-positive |
Capital per unit of capacity is 3x to 8x more efficient in the microfactory model, and the comparison actually understates the advantage because Katerra's capacity was theoretical while Reframe has homes in production at multiple sites. A microfactory that fails loses months and single-digit millions, not years and billions, and that failure tolerance is what lets you iterate, learning from the Somerville factory and applying those lessons in California within weeks rather than the multi-year feedback loops that centralized megafactories impose.
What Still Has to Go Right
Fifty homes per year per factory is a rounding error against national housing demand. The US needs roughly 4 million additional housing units to close the current shortage, according to Freddie Mac and the National Association of Realtors. At 50 homes per factory, you need 80,000 microfactories running simultaneously to close that gap. Nobody is claiming microfactories alone solve the housing crisis.
Transportation logistics remain a real constraint because modular components travel on flatbed trucks, and truck dimensions limit what you can build in a factory and assemble on-site. Reframe uses helical pile foundations instead of poured concrete, which eliminates one site-work bottleneck but doesn't eliminate the others: utility connections, grading, landscaping, final inspections. Weather doesn't matter inside the factory, but it still matters outside it, and that last-mile exposure hasn't changed.
Building codes remain fragmented across 50 states and thousands of local jurisdictions, and while Reframe's AI handles code-compliant design generation, the permitting process itself remains manual, slow, and unpredictable. A home that takes 100 days to build in a factory can still take 180 days to permit in a jurisdiction with an understaffed planning department. The software adapts the design to any jurisdiction, but nobody has automated the bureaucrat who stamps the permit.
And the VC timeline mismatch that killed Veev hasn't evaporated. Reframe's investors need returns within a fund lifecycle, typically 7 to 10 years. If microfactories take 15 years to reach meaningful market share, the capital structure breaks before the production model proves itself, exactly as it did for the last generation.
Strongest Counterargument
Modular and factory-built housing has been promising to disrupt construction for three decades. It hasn't. Market share has flatlined below 5% through multiple technology cycles, venture booms, and housing crises. There is a reasonable case that the problem isn't the factory model at all but something deeper about construction as an industry: hyperlocal labor markets, adversarial permitting regimes, consumer preference for custom over standardized, and a skilled trades workforce that views factory production as a threat to be resisted rather than an efficiency to be embraced. FactoryOS's industrial CBA pays factory workers roughly half the hourly rate of field carpenters, which makes the economics work but also explains why much of the existing construction labor force has no incentive to cooperate with the transition. If the structural barriers are cultural and political rather than technical and financial, better factories with smarter software don't help. No algorithm fixes that.
What This Means If You're Building or Buying
If you're a general contractor running $2M to $5M residential projects, the microfactory model is not a theoretical threat to your business right now. Fifty homes per year in Somerville doesn't affect your pipeline in Phoenix. But the model is designed to scale geographically, and if Reframe or its competitors successfully launch 10 to 20 microfactories within five years, you should understand the economics you'll compete against: 35% lower cost, 2.5x faster delivery, and consistent quality that doesn't depend on which framing crew showed up sober on a given Tuesday.
If you're a homebuyer in a supply-constrained market, factory-built homes from companies with actual revenue and actual completed projects are worth investigating. Azure's Welcome Home Village opens June 8 in San Luis Obispo. Reframe has 20 units under construction in Somerville and 12 single-family homes in Devens, Massachusetts. These are real addresses with real construction timelines, not renderings and press releases. Check for yourself.
If you're evaluating these companies as an investor, the question is not whether AI-powered microfactories produce better homes, because they probably do, but whether 50-home-per-year factories can reach the volume where unit economics improve fast enough to generate venture-scale returns before the fund clock runs out. Azure's positive EBITDA at $7 million in revenue suggests the unit economics work. Growth is the question. Whether the growth economics work at the speed venture capital requires is the $3 billion question that Katerra, Veev, and Skender answered in the negative.
Limitations
Katerra's actual peak production numbers are not publicly available, so the 2,000 to 4,000 unit annual capacity estimate is derived from reported facility size, comparable factory output, and staffing levels at peak employment. Reframe's cost reduction claims (35% cheaper, 2.5x faster) are company-reported and have not been independently audited. Azure's revenue and EBITDA figures come from the company's own press releases via GlobeNewsWire. FactoryOS wage comparisons are sourced from HBR reporting and may not reflect current contract terms. No site visits were conducted for this article, and all claims are sourced from published reports, regulatory filings, and industry surveys. National housing shortage estimates from Freddie Mac and NAR use different methodologies and produce different figures; 4 million is a rough consensus midpoint, not a precise count. This article does not evaluate financing options, warranty terms, or long-term maintenance costs for factory-built homes, any of which could materially change the buyer calculus.