By Sarah Xiao 
American manufacturers have been told for a decade that automation pays for itself. The problem was never the payback — it was the balance sheet in between. Formic’s business exists in that gap, and this week it raised $50 million to widen what fits through it.
What Formic actually sells
Chicago-based Formic, founded in 2020, sells a Robot-as-a-Service contract: equipment, deployment software, field operations, maintenance, financing and performance monitoring bundled into a recurring monthly payment, with no upfront capital outlay. The manufacturer does not buy a robot. It buys a working cell, and a vendor whose revenue depends on that cell still running next quarter.
The announced figures, which the company reports and which no independent audit has yet confirmed:
- $50 million in growth capital — a Series B led by Mercato Partners, with Light Ray Ventures, Lightscape Partners, Citi and existing investors, plus debt from S2G Investments. Formic has not disclosed the equity/debt split.
- Hundreds of robots at more than 100 facilities across over 30 states.
- 750,000+ hours of live production — roughly 86 years of continuous runtime, spread across a fleet.
- 99.6% fleet uptime and 97% customer renewal.
- Deployment roughly three times faster than a year ago.
- Assembly and test in Bolingbrook, Illinois; R&D in Oakland, California.
Applications today are the unglamorous ones that actually pay: palletizing, case packing, machine tending, kitting and assist work across food and beverage, CPG, chemicals, aerospace, defence and packaging.
The 750,000-hour number is the real asset
Funding announcements are usually about money. This one is about an operating record, and it is worth separating the two.
750,000 hours is not a demo reel. It is the accumulated evidence of what happens when a robot meets a real factory: parts change, shifts change, fixtures wear, conveyors jam, a customer changes a carton. Every one of those events is a corrective action, and every corrective action is a piece of knowledge that a laboratory cannot generate.
That record does three things at once, and only one of them is about AI:
- It prices risk. With 750,000 hours and 99.6% uptime, Formic can quote a monthly fee instead of a capital price, because it can estimate the downtime distribution. A vendor with 5,000 hours cannot.
- It compresses deployment. Three-times-faster deployment is not a software achievement; it is a playbook achievement. The second hundred installations are cheaper than the first because the failures are already known.
- It becomes training data. Formic is working with Physical Intelligence to adapt that company’s robot foundation models using its own production data — a relationship that predates this round and appeared in Physical Intelligence’s October API beta.
The strategic claim underneath all this is Mercato’s Ryan Sanders framing: as robots get smarter, the bottleneck moves from the lab to the loading dock. Deploying, servicing and financing robots in real plants is a different business from building them.
The arithmetic a buyer should run
RaaS is not automatically cheaper. It is differently shaped, and the shape matters.
What the buyer gives up: the asset, the residual value, the option to redeploy the machine to another line, and the ability to stop paying by stopping production. A monthly fee is a fixed cost; a purchased robot is a sunk one.
What the buyer gains: no capex approval, no maintenance organisation, no integration risk carried internally, and — the underrated part — a vendor with an incentive to keep the line running. Under a purchase model, the integrator is paid at acceptance and leaves. Under RaaS, the integrator is paid every month and only gets paid if the cell produces.
That changes which number the contract is written around. In a purchase, the negotiation is about price and cycle time. In RaaS, it must be about availability and what counts as a billable hour. Four questions decide whether the deal is good:
- Is the fee per hour of runtime, per month, or per unit of output? Per-hour billing rewards the vendor for keeping the machine busy; per-output billing puts yield risk on the vendor. They are not interchangeable.
- What is excluded from “uptime”? Planned maintenance, changeovers, customer-caused stoppages and material starvation are usually carved out. Read the carve-outs, not the headline percentage.
- Who pays for parts and wear items? Grippers, suction cups, belts and fixtures are consumables with real annual cost.
- What happens at contract end? Buy-out price, removal, and whether the deployment data stays with the plant.
A vendor quoting 99.6% uptime is quoting an average across a fleet. Your cell will not be the average. Ask for the number on your application class, over the last four quarters.
Where this connects to the humanoid story
The same week, two other signals landed that look unrelated and are not.
AGCO announced a humanoid pilot at its Hesston, Kansas plant from 2027 — AI-driven automation for repetitive tasks. It is a plan, not a deployment: no scale, no ROI results published. What makes it interesting is the setting. Automotive pilots run high volume and fixed takt. Agricultural equipment runs multi-variant, low-volume, large-part assembly, with heavier workpieces and more task variation. A humanoid that survives there has demonstrated something a car-plant video does not.
Agibot (智元) shipped two pieces of infrastructure in two days — a fourth embodied-reasoning dataset with 150 hours of real robot data, and Genie Studio 2.0, a platform aimed at standardised delivery and replication. The distinction the company draws is the right one: grasping solves motion; reasoning solves what to do next, and what to do when it goes wrong. A factory pays for the second.
Put the three together and the shape of the next phase is visible. The scarce asset is no longer a robot. It is the recorded experience of a robot doing paid work — Formic’s hours, Agibot’s datasets, AGCO’s pilot data. Whoever accumulates it can price risk, and whoever can price risk can sell automation as a service rather than as a machine.
Two things to be sceptical about
The fleet numbers are vendor-reported. 99.6% uptime, 97% renewal, one in two Americans having used a product their robots packed — these are company claims in a funding announcement, not audited figures. They are plausible for a mature palletizing fleet and they are still claims.
General-purpose is a roadmap, not a product. The move from task-specific cells to general-purpose robots is the stated use of the funds. Task-specific automation is a solved commercial problem with known unit economics; general-purpose is not. The transition is where the risk sits, and the honest read is that Formic is betting its operating record can be converted into model performance faster than the model companies can build their own deployment arms.
The three questions that travel
The source article ends with a reader’s checklist, and it is a good one — worth keeping because it applies to every robotics announcement, in any language:
- Is this robot working continuously on a real production line, or performing on a stage?
- How are purchase, rental, maintenance and downtime costs accounted for — and is the low number the entry cost or the total cost?
- Are there real orders, repeat customers and stable operating data — and are plans clearly separated from things that have already happened?
Add a fourth, which this week makes unavoidable: check the date before you check the video. A widely shared clip of a Figure 02 humanoid “retiring” by jumping into molten steel circulated again this week; it was three weeks old. Footage looking real does not make the news current.
Related reading
- The invisible hours: what robot commissioning really costs — the labour RaaS is quietly absorbing.
- Robot cell three sums: what to check before you buy — the same arithmetic, from the buyer’s side.
- Tesla’s Optimus line: 1M a year, 20k a week — why “off the line” is not “in production”.
- Dream Labs: robot intelligence at $3 an hour — the other per-hour model now being sold to factories.
- The Big Four at 38%: hardware commoditises, engineering does not — why the value is moving to deployment.
Sources: the $50 million round, the Mercato Partners lead, the S2G debt facility, the fleet figures (100+ facilities, 30+ states, 750,000+ production hours, 99.6% uptime, 97% renewal, 3× faster deployment) and the Physical Intelligence collaboration are from Formic’s announcement of 22–23 October 2026, as reported by Industrial Equipment News, The Next Web, Axios, AI Insider and Robotico. The AGCO pilot, Agibot AI WAVE releases, the Tesla status note and the Figure 02 retraction are as compiled by 机械行业观察 on 24 October 2026. The RaaS cost arithmetic, the four contract questions, the reading of the 750,000-hour record and the scepticism notes are Latest Robot News editorial. Image: 机械行业观察.