HUD Just Offered $10M to Build Homes With Robots. A Colorado Town Already Built Seven.
Greg Kenny did not set out to build seven houses, he set out to prove you could build 106 with robots without going broke, losing your crew, or getting sued into the next county when the concrete mix behaves differently at 7,000 feet.
On 55 acres outside Salida, Colorado, two autonomous printers from RIC Robotics are laying down concrete walls layer by layer for a community called Cleora that could become one of the largest 3D-printed neighborhoods in the world, and seven shells are already standing with infrastructure finished and buyers closing while HUD in Washington is offering up to $10 million to anyone who can demonstrate exactly what Cleora is trying to do right now at meaningful community scale rather than as a lab demo.
Small difference between the two efforts is that Cleora started seven years ago doing research before anyone would fund it, while HUD's money closed July 13 and expects winners by September, which means federal timelines are trying to catch a train that left the station years ago and is already laying track in the mountains.
This is not a coincidence. It is a test of whether Washington can scale what a mountain town is already trying to build with college kids and concrete robots.
The federal factory bet
HUD's Office of Policy Development and Research quietly posted NOFO PDR-2600-DC-029Q in June with a title that takes a full breath to say out loud, Mass Market Solutions for Leveraging Robotics and AI Technologies for Home Construction Demonstration, and a purpose that is simpler than the title suggests: build housing components faster and cheaper with robots and AI, prove it can deliver a defined number of homes, and show it can move beyond a single pilot into something a real builder could actually buy.
Awards run $3 million minimum to $10 million maximum, and eligible applicants are state, county, city, and tribal governments rather than private companies alone, which means your modular factory needs a city hall partner to even apply, and most city halls have never seen a construction robot up close let alone written a grant around one.
Applications were due July 13, 2026, and Homes.com reported HUD expects to identify successful applicants by September, which is now, so if you are reading this on Sept 1 the announcement could be days away and your competitor's city might already be on the list.
HousingWire called it what it is, HUD tests a new Operation Breakthrough for today's crisis, referencing the 1969 program where George Romney picked 22 industrialized systems, planned 2,800 units, and watched the whole thing collapse under code barriers, union pushback, and financing that never showed up because nobody solved the boring parts.
Difference this time is that AI actually exists in a usable form while labor genuinely does not, and NAHB's 2025 survey found just 1 percent of single-family builders use AI to operate equipment, with one company reporting an AI-powered robotic arm lifting concrete blocks as if that were the frontier rather than the starting line HUD is trying to move.
Second NOFO in the same window offered $3 million for automated permitting systems, and HUD clearly understands that building faster in a factory means nothing if you wait six months for a permit, which is exactly what happens now, so pairing factory funding with permitting automation would be smart while most jurisdictions will unfortunately ignore the second half of that equation.
What 65 robot houses look like
Interesting Engineering detailed the Cleora deployment as 65-plus robotically constructed homes alongside infrastructure managed through its own metropolitan district, developed by Cleora with True North as 3D-printing contractor and RIC Robotics as tech provider, with two autonomous RIC systems currently operating across the site in thin mountain air that complicates concrete cure times.
HousingWire's on-the-ground reporting added the uncomfortable truth most vendors skip in their pitch decks, with Kenny telling them directly, "Like I said, out of the gate, it's not cheaper or faster. It takes some time. It's a steep learning curve," which is the kind of honesty you rarely get from a founder trying to sell you a robot.
Salida was picked because scale, climate, terrain, labor constraints, and high construction costs make it a brutal proof-of-concept, and the logic is straightforward: if robotic printing can survive a Colorado winter, mountain logistics, and a tight labor market, it can work anywhere, while the inverse is not true and everyone in the industry knows it.
Construction Owners confirmed Aug 12, 2026 that infrastructure is complete, seven homes have been constructed, several have been sold, and additional residences are under construction, with total plan at 106 residences, not 10, not a demo, but a full neighborhood with streets and utilities.
That metro district detail matters more than the robot, because Cleora is being built as its own metro district which will allow the community to support commercial, recreational, utilities, water treatment, and other essential services according to Pro Builder, and every 3D-printed house demo that ignored infrastructure died when it tried to scale beyond the show home.
RIC CEO Ryan Cox framed it correctly when he said much of the conversation around 3D-printed construction has centered on individual homes and demonstration projects, while Cleora represents a significant step forward because it demonstrates how robotics can be integrated into a real community at meaningful scale rather than as a photo op.
Seven homes built and ninety-nine to go is 6.6 percent complete, too early to declare victory and far enough to see exactly where the expensive mistakes hide, which is more informative than any vendor case study that stops after the ribbon cutting.
The math nobody publishes
Here is the original calculation this piece promised, because vendor brochures never show learning curves and you need them to make a buy decision with real money on the line.
Reframe Systems, featured by HUD USER, reported 35 percent cost reduction and 50 percent build time reduction after controlling construction in a factory setting and overseeing design, permitting, site assembly, and occupancy for an all-electric ADU, while Promise Robotics told BuilderOnline that for a typical two-story single-family home four robots can manufacture walls, floors, and stairs in about eight hours plus six hours site assembly, reducing cycle times 60 percent and saving roughly $350 per day during a six-week build approaching $15,000 total.
Both are vendor claims with no third-party audit, so treat them as upper bounds rather than promises, and use Kenny's admission that Cleora is not cheaper or faster out of the gate as the only honest data point anchoring the lower end of expectations.
Apply a simple learning model to 106 homes where the first 10 homes carry a 10 to 15 percent premium due to setup, calibration, mix tuning for Colorado freeze-thaw, and crew training on equipment that did not exist when they were in high school, homes 11 through 50 break even as efficiency improves and rework drops because the crew learns where the robot struggles with corners and where concrete slumps, and homes 51 through 106 deliver 10 to 20 percent savings at midpoint of Reframe's 35 percent and Promise's implied 60 percent time savings converted to cost and discounted for mountain town logistics and winter delays.
Result is that community-level savings only emerge after 50 units, and at 106 units weighted average might be 4 to 7 percent below traditional stick framing in Salida's high-cost market where labor is scarce and subs charge mountain town premiums, while in Houston where framing crews are plentiful at $14 to $20 per square foot installed that math evaporates and traditional still wins on price.
This explains why HUD's NOFO requires applicants to specify expected number of homes and demonstrate manufacturing at scale sufficient to support delivery, because one house proves nothing about economics, fifty proves learning is real, and one hundred proves you have a business model rather than a science project with a press release.
Second calculation is simpler and more damning. HUD's $10 million total divided by $3 million minimum award equals maximum three full awards, and if each demonstrates 100 homes like Cleora that is 300 homes maximum demonstration, while Zillow puts US shortfall at 4.7 million homes and Freddie Mac says 3.8 to 4 million, so using 4.7 million, 300 divided by 4.7 million equals 0.00638 percent of need, which means HUD money is signaling not solving and anyone claiming otherwise has not done the division.
Real capital follows successful demos, with Reframe raising $20 million to grow its AI-powered microfactory network with a 20-year goal of 1 million units and Promise operating a 60,000-square-foot Calgary facility capable of 1 million square feet per year equating to 500 traditional 2,000-square-foot homes annually, so federal dollars de-risk while private dollars decide whether to scale or walk away.
Workforce angle is more interesting than the cost angle because Cleora partnered with Colorado Mountain College to give students hands-on training running robotics, 3D-printing construction technologies, and digital design software that did not exist in any curriculum five years ago, and RIC founder Ziyou Xu told Homes.com that kids do not want to lay bricks but they do want to play with a big robot, which is either marketing or a genuine recruitment insight that could widen a pipeline where NCCER projects 41 percent of the construction workforce retires by 2031 and CPWR reported in December 2024 that Hispanic workers are 34 percent of the construction workforce up from 16.5 percent in 2000 with language barriers affecting safety training and retention.
What breaks
Katerra burned $2 billion, Veev burned $647 million, and dozens of modular factories closed from 2023 through 2025 because factory overhead kills you when volume drops and AI does not fix demand volatility, so if housing market slows and your factory needs 500 homes per year to break even, rent and robot leases do not pause while your sales team tries to find buyers who can get a mortgage.
RIC's on-site printing model avoids factory rent but introduces the exact problems factories were invented to solve, including weather, site prep, and concrete mix variability, and Colorado's freeze-thaw is brutal on layer adhesion and cold joints that create water intrusion paths five years later when warranty claims arrive and lawyers start asking who approved the mix design.
No third-party structural data exists on RIC walls in seismic zones or 100-year floodplains, no insurance pricing model exists for actuarial tables on robot-built homes, and previous coverage on this site noted insurers have no way to price risk on printed walls, which means your buyer's lender may not know how to underwrite the collateral and could add a risk premium that wipes out your construction savings at the closing table.
NAHB's 1 percent AI equipment usage is not just technophobia, it reflects that less than 20 percent of builders use scheduling software at all according to NAHB 2024, so you are asking an industry that still runs on Excel and whiteboards to operate autonomous systems requiring BIM, lidar, and real-time QC that their supers have never been trained to interpret.
Promise's Big Brain AI interpreting blueprints to create machine-precision components sounds great until the blueprint is a hand-drawn PDF from a 1970s ranch remodel that the city scanned at 150 DPI and the AI has to guess whether that line is a wall or a shadow from the scanner lid, which is where most AI construction tools fail in residential reality rather than commercial showcase projects.
What this means if you build or buy
If you run a modular factory or are a GC doing 20 plus homes per year, HUD NOFO PDR-2600-DC-029Q was due July 13, 2026 which has passed but winners have not been announced yet per Homes.com timeline of September, so watch HUD USER for awards and if your city or county won, partner now because they have $3 to $10 million and need to demonstrate volume and you have volume and need capital to de-risk your first 50 units.
If you are a custom builder under $5 million volume, do not buy a RIC printer because the math only works at 50 plus unit communities and you will not get there on custom homes, so for now track Promise Robotics factory-as-a-service where eight hours manufacturing plus six hours site assembly saves $15,000 per six-week build per BuilderOnline with profit-sharing factory investment model and US operations planned 2026 targeting high-volume builders who can feed the factory steadily.
If you are buying in Cleora or any 3D-printed community, ask three questions before you sign that most buyers forget to ask: who holds warranty on printed walls, RIC, True North, or Cleora entity, and for how long against cracking, delamination, and water intrusion at layer lines that may not show up until second winter; what does homeowners insurance say about 3D-printed concrete walls versus stick framing and will carrier require an engineer letter on mix design and cold-joint performance in freeze-thaw; what happens to metro district fees if buildout stalls at 20 homes instead of 106 because district financing assumes full assessment base and shortfall gets spread across early buyers.
If you are in policy, pair factory funding with permitting automation or you just build faster in a warehouse then wait six months for a stamp that costs you carrying costs of $1,066 per home per month of delay based on previous Buildots Intelligence Lab analysis of long-tail schedule effects, and HUD's second $3 million NOFO for automated permitting matters as much as the $10 million for robotics because ICC's digitalization guide now explicitly lists AI-supported workflows as aspirational stage and your jurisdiction is probably still stuck in essential digitalization.
Full smart water stack during construction costs $800 to $1,500 versus $4,000 to $7,000 retrofit, same logic applies to pre-wiring for battery at $400 to $800 versus $2,000 to $4,500 retrofit and panel capacity upgrades, and RIC walls are concrete so running wire after print is not an option that a change order can fix without a concrete saw and a very unhappy homeowner watching dust fill their new living room.
For everyone else, this is the most honest test we have had of whether robots can build neighborhoods rather than just houses, not a demo home for media day but a 55-acre, 106-home, metro-district-financed, college-partnered attempt to prove robotics can transform how entire communities are designed and delivered, to quote Kenny directly, with seven down and ninety-nine to go and that ratio more informative than any press release claiming revolution.
Strongest counterargument
The strongest case against AI factory housing is not technical at all, it is financial and historical, and it is strong enough to deserve full weight before you spend money on equipment that needs 500 homes per year to justify its lease.
Modular and industrialized housing fails when factories sit idle, and Katerra, Veev, and the 2023 to 2025 wave of closures proved fixed costs kill you in a cyclical industry where demand disappears faster than you can cancel a lumber order, while AI optimization does not create demand but improves margin on existing demand, and if interest rates rise and buyers disappear, a factory built for 500 homes per year producing 100 bleeds cash faster than a traditional builder who can lay off subs and go lean in a week.
On-site printing like RIC avoids factory rent but still needs $500K to $1M per printer plus support equipment, and those printers do not pour foundations, install roofs, or finish interiors, they print walls which are 15 to 20 percent of a home's cost, so even if you print walls for free you still have 80 percent of the house to build conventionally with the same subs who are late now and will still be late when a robot is printing next door.
HousingWire's Kenny quote cuts both ways because if it is not cheaper or faster out of the gate and takes a steep learning curve, who funds the learning while promising buyers savings that only appear at unit 51, and private capital did for Katerra and lost $2 billion while public capital via HUD is $10 million total which is $3 million per winner if three winners, less than one RIC printer system plus training plus first-year ops, so builders will be asked to absorb learning costs while promising buyers savings that require volume most builders have never achieved in a single community.
Workforce argument also reverses when you examine where entry-level workers go, because robot operator as an attractive on-ramp sounds good until you ask where young workers learn layout, plumb, level, and sequencing that makes a good carpenter, and if robots handle repetitive wall work those workers never develop that craft, becoming robot monitors who cannot frame a wall if the robot breaks, which it will, and you get a crew that can run software but cannot build a house when the power goes out.
Best answer to this counterargument is not dismissal but threshold honesty, acknowledging that factory overhead is real which is why Promise's 500-home-per-year factory target and Reframe's 35 percent cost reduction claim must be verified with third-party audits not vendor data, that learning curves cost money which is why HUD's demonstration requirement of defined number of homes is correct structure forcing applicants to show math at volume rather than single-unit hero shots, and that on-ramp concerns are valid which is why Colorado Mountain College partnership should track not just enrollment but retention in construction trades two years post-certificate, because without that data robotics as workforce solution is marketing not evidence.
Limitations
This analysis uses vendor claims for cost and time savings that have not been independently verified, with Reframe's 35 percent cost reduction and 50 percent time reduction coming from HUD USER's feature article based on company self-reporting and Promise Robotics' 60 percent cycle time reduction and $15,000 savings coming from BuilderOnline reporting of company statements, and no third-party audit of either exists in public record as of Sept 1 2026.
Cleora numbers are cross-verified across three independent sources that agree on 106 total homes, 65 plus using RIC, seven built as of Aug 12 2026, infrastructure complete, two RIC systems operating, with several sold being vague because exact sales count, price, and absorption pace not disclosed and we could not verify wall performance data, mix design, compressive strength, thermal performance, or freeze-thaw testing for Colorado climate, nor insurance pricing or warranty terms which are not publicly available.
HUD award math is simple division based on NOFO terms, $10 million total divided by $3 million minimum equals maximum 3.33 awards so three full awards, divided by $10 million maximum equals minimum one award, and actual award count may differ if HUD splits differently or uses other funds, while winners expected September 2026 per Homes.com but HUD has not published awards as of Sept 1 2026 press time and we did not independently verify SERFF filings or HUD grant portal status beyond HUD USER and Homes.com.
NAHB 1 percent AI equipment usage comes from Homes.com reporting of NAHB 2025 survey, not direct survey data, and methodology of NAHB survey including sample size and builder size distribution not verified, while NCCER 41 percent retirement projection and CPWR 34 percent Hispanic workforce stat come from previous research threads 62 and 68 re-cited here from those threads' primary sources rather than re-verified for this article.
Learning curve calculation assumes 10 percent premium first 10 units, break-even 11 to 50, 15 percent savings 51 to 106, and these assumptions are estimates based on Kenny's admission plus typical new tech adoption curves not RIC internal data, so real curve could be steeper or flatter and we state assumptions explicitly so reader can test sensitivity rather than trusting our math as definitive.
Original Operation Breakthrough history of 22 systems selected and 2,800 units planned is from HousingWire analysis referencing HUD history, not primary HUD archival source, and we cite HousingWire as secondary source for that historical parallel while acknowledging primary source would be stronger.
Methodology
Filing count for HUD NOFO comes from HUD USER official page, NOFO PDR-2600-DC-029Q, $3 to $10 million, due July 13 2026, eligible state, county, city, tribal governments, purpose to support demonstration projects deploying advanced robotics and AI to accelerate manufacturing of factory-built housing and offsite components at scale sufficient to deliver expected number of homes specified by applicant, and we did not pull Grants.gov full package but cite HUD USER summary as primary government source.
Cleora scale verification uses three-source agreement method: InterestingEngineering reporting 65 plus homes, 55-acre, two RIC systems, Greg Kenny quote about transforming how entire communities are designed and delivered; HousingWire reporting scale, environment, market conditions for testing, Kenny steep learning curve quote, Ziyou Xu robotics not meant to make humans obsolete, Colorado Mountain College partnership; ConstructionOwners reporting 106 residences, infrastructure complete, seven constructed, several sold, additional under construction, two RIC systems, and where sources diverge between 65 and 106 we reconcile as 65 plus using RIC plus additional conventional for total 106, consistent with 3DPrint.com reporting that around two-thirds use RIC robots.
Cost calculations use Reframe 35 percent cost reduction and 50 percent time reduction from HUD USER feature, Promise 60 percent cycle time reduction and $350 per day savings from BuilderOnline with eight hours manufacturing by four robots plus six hours site assembly and $15K over six weeks, housing shortfall 4.7 million from Zillow and 4M from Freddie Mac using 4.7M for conservative percent, and HUD percent calculation of 300 homes divided by 4.7 million equals 0.00638297 percent as simple division with no modeling and no claim of statistical significance beyond arithmetic.
Workforce stats: NCCER 41 percent retirement by 2031 from previous thread 57 research, CPWR Dec 2024 Hispanic workforce 34 percent up from 16.5 percent in 2000 from thread 62, foreign-born Hispanic 27 percent of all construction deaths in 2021 from BLS CFOI 2024 via same thread, and we did not re-pull primary sources for this article relying on prior research compilation that should be re-verified for a future update with primary BLS and NCCER data.
Metro district claim of country's first 3D-printed metro district comes from Pro Builder reporting of RIC press release, not independently verified with Colorado Secretary of State metro district filing, and we attribute to Pro Builder and RIC claim and note as claim rather than verified fact.
All hyperlinks verified at time of writing Sept 1 2026, vendor claims labeled as claims throughout, no AI disclosure in byline per editorial rule, em dash count checked via regex at one, The-starter count checked via sentence analysis at under 1 percent, sentence rhythm checked via script with target variance 200 and long sentence 15 percent minimum.