Direct answer
1872 is the small-plant case. Three engineers from SpaceX and GE Aviation opened a steel fabrication shop in Cincinnati’s Camp Washington on 22 July 2026, with a $15 million seed, among the largest in Ohio, from funds advised by The O.H.I.O. Fund. They did not sell scheduling software to existing fabricators. They built Factory One and run the jobs themselves.
Dan Summers, the chief executive, has said the other half of the idea was the power of putting software, hardware, and product development on one clock, learned at SpaceX. The Conveyor put the make-or-buy call in one sentence: instead of selling software to fabricators, 1872 is building the plant.
The stack they describe is Factory OS, split into Architect and Conductor. Architect takes the three-dimensional design and plans, prices, and sources the job. Conductor runs the floor. It is meant to drive machines, not only report on them. Agents are trained as stand-ins for buyers, planners, welders, and logistics. Weld cells come from Columbus-based Path Robotics, including the Obsidian model. The first product focus is steel skids, rectangular frames, aimed at data-center and small-modular-reactor builders. The lead times they want are months down to weeks. The full-autonomy speech is 2027. Summers told Ars Technica they may stop at 80% if the last twenty points do not pay.
PitchBook lists nine people and a Cincinnati address at 2629 Spring Grove Avenue, the 1903 David Hummel Building Company site. Customers exist, Summers told ENR, and will stay unnamed until the plant is closer to the 2027 target. Treat that as a seed-stage fact, not as a hidden large-customer book.
xvertical does not work for 1872. 1872 is not a client. The company is here only as a case a reader can check. Brian Mongilio, named below, is a founder of 1872. He is not the Brian who signs this site.
Where this sits next to the other cases
| Case | Scale | Owned layer |
|---|---|---|
| BYD | A global carmaker | Cell to car |
| Anduril | A defense product firm, with a factory program | Lattice, plus enough factory |
| Harper, AllDigital, OffDeal | Service firms | Policy, bind, and close |
| 1872 | $15 million, about nine people, one building | Design to weldment in one shop |
1872 is what “the factory is the product” looks like before a large government contract. It is also what happens when a Path Robotics weld cell, a tool sold to other shops, sits inside a firm that owns the job. Crosby is a comparison used elsewhere in these essays. There is no Crosby essay on this site.
What vertical means in Camp Washington
A traditional structural-steel job:
- The customer’s drawings
- Estimating and the material buy
- Nesting, cutting, and fit-up
- Weld
- Check and ship
- The schedule that connects steps 2 to 5, usually a person and a whiteboard
A robot vendor owns step 4. A scheduling vendor owns a report on step 6. 1872 claims steps 1 to 6 on one system, in one building, on jobs it is paid to deliver.
Factory OS is the command layer of this story, with a smaller vocabulary:
- Architect reads the model and writes the plan: routing, the buy, and the price.
- Conductor is supposed to execute: move steel, assign cells, and run the shift without a separate automation project bolted onto a manual line.
Summers’ complaint about the rest of the industry, to ENR: shops drop robots into manual workflows and underuse them. The integration is not the arm. It is whether Conductor can sequence the arm, the crane, and the next kit without a coordinator walking the bay.
Path Robotics is rented capability, the way a chip designer rents a foundry. Obsidian welds. 1872 sequences the work. Mark Kvamme of the O.H.I.O. Fund, an early Path backer, and Path chief executive Andy Lonsberry are part of the origin story: islands of autonomous welding, and no flow from the dock to the dock. 1872 is the attempt to connect the islands and be the customer of that flow.
The name 1872 is the year of Bessemer’s industrial moment in the company’s telling. That is marketing. The operational claim is 2026 and 2027.
The numbers that matter
| Claim | Figure | Source | Date |
|---|---|---|---|
| Founded | 2025 | PitchBook | — |
| Opened Factory One | 22 July 2026, Camp Washington | Company / AP | 2026 |
| Address | 2629 Spring Grove Avenue, the 1903 Hummel building | PitchBook / AP | — |
| Founders | Dan Summers (chief executive), Brian Mongilio, Michael Grant, from SpaceX, with GE Aviation in the mix | ENR / The Fabricator | 2026 |
| Seed | $15 million, through vehicles advised by The O.H.I.O. Fund | Company | July 2026 |
| Headcount | 9 | PitchBook | 2026 |
| Operating system | Factory OS, Architect plus Conductor | Summers to The Fabricator / Ars Technica | 2026 |
| Weld partner | Path Robotics / Obsidian | Company | 2026 |
| First product | Steel skids and complex weldments for energy, utilities, and industry; data centers and small modular reactors named | Ars Technica / PitchBook | 2026 |
| Lead-time goal | Months down to weeks | Company | 2026 |
| Autonomy goal | Most of the process by 2027; a possible stop near 80% | Summers to Ars Technica | 2026 |
| Named customers | None public | Summers to ENR | August 2026 |
| Path first-pass weld yield, a vendor claim | 95 to 100%; about 70% arc-on time | Ars Technica citing Path | 2026 |
How to read the table. The building, the raise, the system names, and the Path partnership are solid. Throughput, yield on 1872’s own jobs, and “first autonomous steel shop in the United States” are company belief. Path’s cell metrics are Path’s, not a teardown of Factory One. Nine people and unnamed customers means this is a design with a ribbon-cutting, not a finished production record.
Why this is a plant, not robots in a job shop
They kept the job. The same sentence as Harper and OffDeal. Software into a fragmented, short-staffed trade dies in the estimator’s inbox. Owning the weldment lets Architect and Conductor train on real kits.
The factory is the first customer of the system. You do not learn dock-to-dock flow from a slide about other people’s bays.
They will not insist on 100% autonomy. Summers’ 80% ceiling is the adult sentence in the coverage. Lights-out is a fundraise line. Knowing where a person still beats the cell on a complex fit-up is operations.
SpaceX is a method reference, not a product reference. Summers does not describe SpaceX as a rocket company in this coverage. He describes it as product development plus manufacturing. xvertical’s SpaceX essay is the public case of that method. 1872 is a founder’s prior, not a sequel.
What 1872 still does not own
- The weld model. Obsidian is Path’s. 1872 owns the sequence, not every instrument.
- A published production rate. No tons per week, no first-pass yield on their skids, and no named data-center job.
-
- The plant opened in July 2026. Autonomy is a date, not a current state.
- The category “American manufacturing.” One bay in Camp Washington. Expansion talk is talk.
- Proof that Architect and Conductor beat a sharp estimator plus two Path cells in a conventional shop. That comparison is the one that matters, and it is not public.
- Independence from one regional fund and one robot maker. Fine at seed. A concentration risk later.
Same four-part test
| Test | BYD and Anduril | 1872 |
|---|---|---|
| Expensive layer is scarce | Cell, airframe, command software | Skilled weld hours, and the dock-to-dock sequence |
| Volume amortizes the stack | Millions of cars, thousands of airframes | Repeatable skids and frames, if the book appears |
| Model co-designed with the machine | Driving chip, command software on the machine | Architect on the design, Conductor on the cell |
| You are the first customer | Own fleet, or an Army division | Factory One is the first user of Factory OS |
It fails if 1872 becomes a Path reseller with a nicer dashboard. It also fails if they chase 100% autonomy past the point Summers already flagged.
What operators should copy
- If the constraint is the bay, own the bay. Do not sell the scheduler.
- Split plan and run. Architect plans. Conductor runs. Do not build one module named “AI.”
- Rent the cell that already welds. Own the sequence that keeps the cell busy.
- Name the first product. Here, that product is skids. A general factory at nine people is a science project.
- Write down the autonomy percentage you will not chase.
Do not copy “largest seed in Ohio” as a strategy. Copy the refusal to drop a robot onto a manual line and call it a transformation.
Key takeaways
- 1872 is plant-vertical integration at seed scale: own the shop, own Factory OS, rent Path’s weld cells.
- $15 million, a July 2026 opening, about nine people, in Camp Washington. Customers are unnamed. Autonomy is aimed at 2027, and maybe only 80%.
- Summers’ own frame: SpaceX taught software, hardware, and product on one clock. Much of steel fabrication still bolts robots onto whiteboards.
- Harper, AllDigital, OffDeal, BYD, and Anduril are on this site. Crosby is not.
Sources
- AP / Business Wire, 1872 launch and $15 million seed, 22 July 2026.
- ENR, “Steel Fabricator 1872 Aims to Integrate AI Into Its Processes,” 19 August 2026.
- The Fabricator, Factory OS as Architect plus Conductor, and the Kvamme / Path origin, 22 September 2026.
- Ars Technica, skids, the 80% autonomy ceiling, and Path yield claims, August 2026.
- Ohio Tech News / O.H.I.O. Fund note, July 2026.
- PitchBook, 1872 profile (employees, address, founding year).
- The Conveyor, “instead of selling software to fabricators,” 27 July 2026.
Lead-time cuts, autonomy dates, “first in the United States,” and unnamed customers are company statements. Path cell metrics are the vendor’s, not a published 1872 production report.
Questions this essay answers
What is 1872?
A Cincinnati steel fabricator built around software. It takes a computer-aided design, plans the job in Factory OS, and welds structural assemblies, first steel skids for industrial and energy customers, in its own shop.
Is 1872 a robotics company?
No. Path Robotics supplies weld cells. 1872 is the shop and the layer that sequences the work.
Why does that count as vertical integration?
Because 1872 owns estimating, planning, floor execution, and the shipped weldment. A robot vendor owns the arc. A scheduling vendor owns the report.
How much has 1872 raised?
$15 million seed, announced with the July 2026 opening, through vehicles advised by The O.H.I.O. Fund.
When will the factory be autonomous?
The company talks about 2027 for most of the process. The chief executive has said they may stop near 80% autonomy.
Who are the customers?
Not named. Dan Summers has said they exist and will be disclosed later.
How is 1872 different from Anduril?
Same idea, different starting size. Anduril has a large Army ordering vehicle and a factory program. 1872 has one building and about nine people. The test is the same.
How is 1872 different from Path Robotics?
Path sells the weld cell, often as a service. 1872 uses the cell and sells the finished steel.