In Pacific Palisades, California, a six-axis ABB IRB 6710 robot arm swings through a pre-programmed arc inside a steel-framed work cell parked on a lot where a family home burned to its foundation eight months ago. It grabs a piece of two-by-six lumber, positions it against a framing jig with millimeter precision, and fastens it into a structural wall panel that will be carried fifty feet to the right and stood up by a crew of four. Twelve weeks from now, a family will move in, and every load-bearing wall in their house will have been fabricated in a structure roughly the size of two shipping containers that arrived on a flatbed truck and will leave when the job is done.
This is not modular housing. Your house doesn't ship from a distant factory on an oversize load escort with pilot cars and DOT permits. Instead, the factory ships to your house.
It is a small but critical inversion of the idea that has been burning venture capital since 2015, and it might be the version that actually works.
Three Machines, Three Bets
Three companies are now operating mobile or rapidly deployable robotic microfactories for residential construction, each with a different machine, a different material strategy, and a different theory about why the last generation of factory-built housing collapsed under its own weight.
Cosmic Buildings, the Los Angeles operation partnered with ABB Robotics, is the most visible. Their mobile microfactory in the Palisades uses ABB's industrial robots and RobotStudio digital twin software integrated with Cosmic's AI-driven building information model, a single digital thread that runs from design through permitting through fabrication. Custom structural wall panels come off the line inside the mobile factory, get delivered just-in-time to the adjacent build site, and go up fast. Cosmic says the approach cuts construction time by 70% and costs by 30% compared to conventional methods in the Los Angeles market. At $550 to $700 per square foot, their homes come in below the $800 to $1,000 per square foot that conventional rebuilds cost in the Palisades, and 45 families have signed on, with a target of 100 homes delivered by 2027.
AUAR, a UK startup backed by ABB's venture arm, operates what they call a "factory in a box," and the name is accurate enough to be useful. Their V2 robotic micro-factory is a single compact work cell that takes sheet timber, cuts it into components, and assembles them into structural elements, all guided by proprietary Master Builder software that generates both the architectural design and the robotic fabrication code from a single model. In March 2026, AUAR completed the first building in the world fully manufactured by a pop-up micro-factory: a two-story structure in Belgium for construction group Vandenbussche NV, with the shell pre-manufactured in under eight hours. Run three micro-factories in parallel and that drops to under three.
Reframe Systems, based in Andover, Massachusetts, was founded in 2022 by three former Amazon Robotics leaders who saw parallels between warehouse logistics bottlenecks and homebuilding inefficiencies. Their 50,000-square-foot microfactories use vision-guided robotic arms, digital interfaces for real-time worker guidance, and production sequencing software that translates a digital home model into staged build instructions for each work cell. Reframe reports 35% cost reductions and 50% time reductions compared to conventional construction, and completed an ADU in Arlington, Massachusetts, in 2024 that became the town's first affordable housing unit of its kind. Since then they have expanded to Southern California for wildfire-zone rebuilds, raised $20 million, and set a 20-year target of one million housing units built through a national network of localized microfactories.
$10 Million in Federal Money, and What It Actually Buys
On June 9, HUD quietly opened applications for the "Mass Market Solutions for Leveraging Robotics and AI Technologies for Home Construction Demonstration," a $10 million program offering awards of at least $3 million each to organizations that can demonstrate AI and robotics in factory-built or offsite housing manufacturing. Applications close July 13.
Separately, HUD allocated $3 million for an automated permitting systems demonstration, testing whether AI can compress the months-long permit review process that stalls projects regardless of how fast you build the physical structure.
Both dollar amounts are modest, and the math is instructive. If HUD funds three microfactory demonstrations at $3 million each, and each demonstration produces ten to fifteen homes, the federal government will have spent roughly $200,000 to $300,000 per data point on whether robots can build houses at scale. Reframe has already raised $20 million privately, and ABB's corporate balance sheet makes $10 million look like a rounding error. Federal money at this scale is not building houses; it is building the evidentiary case that private capital needs to deploy at real scale, because the last time private capital deployed at real scale in factory-built housing, $3.6 billion went up in smoke.
Why Katerra Burned and Microfactories Might Not
Katerra raised $2 billion, opened a 577,000-square-foot factory in Tracy, California, and filed for bankruptcy in 2021 without delivering a sustainable business model. Veev raised $647 million, built a factory in Hayward, California, and collapsed in 2024. Factory OS raised capital, opened in Oakland, and shut down its factory operations.
Every failure shared the same structural flaw: centralized mega-factories required enormous capital expenditure, created logistics nightmares shipping finished modules across state lines, and locked companies into fixed-cost structures that crumbled when demand fluctuated or supply chains broke. A 577,000-square-foot facility that makes homes for one regional market is a bet that one regional market will need a steady stream of factory-built homes at a predictable price for the decades it takes to amortize the facility. That bet keeps losing.
Microfactories invert the economics. Cosmic's mobile unit arrives when there is work and leaves when there is not. AUAR's work cell fits in a shipping container. Reframe's 50,000-square-foot facilities are one-tenth the size of Katerra's factory and designed to serve local markets within a 200-mile radius. None of them are trying to ship finished houses cross-country on flatbed trucks with pilot cars and oversize load permits.
Whether that theoretical advantage survives contact with real-world construction economics at scale remains an open question, because the combined output so far is vanishingly small. Cosmic has delivered zero of its 100-home target. AUAR has completed one building. Reframe has completed one ADU. If you added up every finished structure from the three most prominent mobile microfactory companies in the world, the total would be fewer than a single framing crew in a mid-size production builder's quarterly output.
Only 1% of Builders Care
According to the July 2025 NAHB/Wells Fargo Housing Market Index survey, exactly 1% of single-family home builders use AI to operate construction equipment. Twenty percent use it to write marketing copy. When nonusers were asked how likely they are to adopt AI for equipment operation in the next two years, they gave it a 1.7 out of 5, where 5 means "very likely" and 1 means "not at all." Score 1.7 and you are closer to "absolutely not" than "maybe."
Globally, the picture is similar. RICS's 2025 construction survey found 45% of construction organizations have no AI implementation at all, another 34% are in early pilot phases, and fewer than 1% reported fully embedded organization-wide AI use. Investors, meanwhile, are more enthusiastic: 56% said they planned to increase AI spending. Capital wants in; the people who pour concrete and hang drywall do not.
This is the environment into which HUD is dropping $10 million and three robot companies are parking their factories: a technology that works in the demo, a business model that looks better than Katerra's, and an industry that has given construction-equipment AI a 1.7 out of 5 on the enthusiasm scale.
What a Home Actually Costs When the Factory Shows Up
Seductive numbers need context, because these are market-specific claims being made in exceptional markets. Cosmic's $550 to $700 per square foot sounds cheap until you remember they are building in Pacific Palisades, where $800 to $1,000 per square foot is normal and $500 per square foot would be impossible without imported labor or robots. On a 2,000-square-foot home, Cosmic's approach saves roughly $400,000 to $600,000 per unit at quoted rates, which is transformative for one of the most expensive construction markets in the country but tells you nothing about what happens in Memphis.
Reframe's 35% cost reduction, applied to the national average of roughly $150 to $200 per square foot for construction costs excluding land, yields a 2,000-square-foot home at approximately $195,000 to $260,000 versus $300,000 to $400,000 conventional. A $100,000 to $140,000 savings per home is meaningful, but it is not revolutionary in a market where the median new-home sale price was $495,000 in Q4 2025 (Census Bureau) and land accounts for 20% or more of total cost.
No independent party has audited any of these systems' costs. Cosmic's 70%/30% claims come from Cosmic, Reframe's 35%/50% claims come from Reframe, and ABB's press materials repeat their partner's numbers without independent verification. When ICON claimed 24-hour 3D print times for concrete homes, independent analysis showed the actual construction timeline, including foundations, utilities, and finishing, stretched to months. Microfactory companies will face the same scrutiny when they scale past single-digit unit counts.
What This Means If You Are Building a Home
If you are rebuilding after a wildfire in Los Angeles, Cosmic is taking clients now. If you are a builder in New England, Reframe operates from Andover and is expanding. If you are a European developer, AUAR's V2 micro-factory is shipping. These are not press releases about future products. They are operational systems with real robots assembling real lumber into real wall panels.
If you are a small-to-midsize builder who has been told that factory-built housing requires a $50 million facility investment, the AUAR model challenges that assumption directly. A compact robotic work cell, deployable to the job site, manufacturing the structural shell in eight hours from sheet timber, finished on-site by your existing crews with your existing subs, is a fundamentally different capital commitment than a Katerra-scale factory.
If you are a university or research institution, HUD applications for the $10 million robotics demonstration close July 13, 2026.
And if you are among the 99% of builders who are not using AI for construction equipment and gave the concept a 1.7 on a 5-point scale, consider this: three companies have now parked robot factories on active job sites and started building walls, HUD is spending $10 million to find out if this works at scale, and ABB, a $70 billion industrial conglomerate, has put its name and its robots behind the concept. Between "not interested" and "outcompeted" there is a gap of about twelve weeks and one flatbed truck.
What We Don't Know
All cost and timeline claims come from the companies themselves, and no independent third-party audit of microfactory construction costs, quality, or durability has been published. Combined output from the three most prominent companies is fewer than ten structures. None has operated through a full economic cycle, a material-price shock, or a regulatory crackdown. Cosmic's 70% faster and 30% cheaper numbers are specific to the Palisades rebuild market, where baseline costs run three to five times the national average, and do not transfer directly to markets like Houston or Atlanta where conventional construction is $150 to $200 per square foot. Reframe's Amazon Robotics pedigree is impressive, but warehouse logistics and residential construction are not the same problem, no matter how compelling the analogy sounds in a pitch deck. And the housing deficit numbers that everyone cites, four million to seven million homes depending on methodology, have been used to justify a decade of contech ventures that collectively burned billions without meaningfully bending the supply curve.
Perhaps the strongest argument against microfactories is also the simplest: we have heard this before. Robots were supposed to save construction already, and they did not. Companies betting on robots this time are smart, well-funded, and building real things on real lots for real families who lost their homes. But so was Katerra.
← Back to AI Home Building