Ultra Robotics has packed more than 500,000 e-commerce orders with machines that cannot walk, climb a step, or open a door. The Brooklyn company raised $62 million to build more of them — a $50 million Series A led by Framework Ventures, stacked on a $12 million seed from Y Combinator and NextView Ventures.
Its robot, the OP1 Operator, is a stationary two-armed unit mounted on a conventional industrial arm. The grippers are shaped like forklift prongs. It reaches ten feet up, plugs into a standard 120-volt outlet, and runs up to 24 hours a day.
That description lands in the middle of an argument the robotics industry keeps postponing. Humanoid startups have raised record capital on the promise of a general-purpose machine that fits spaces built for people. Ultra's bet runs the other way — strip the robot down to the tasks that already generate revenue, then sell the outcome by the month.
The humanoid camp
The case for legs is straightforward. Warehouses, factories, and stores were designed around human bodies — the height of a shelf, the reach of an arm, the width of an aisle. A machine shaped like a person drops into that world without rebuilding it.
The money has followed. Humanoid, a UK company, raised $152 million in July at a $1.35 billion valuation. Walden Robotics, a Toyota Research Institute spin-out, launched from stealth the same month with $300 million. Figure AI has raised $1.7 billion in total. In China, UBTECH switched on a factory in Liuzhou built for more than 10,000 humanoids a year, turning out one industrial unit every ten minutes, and XPeng commissioned a line in September that walked its first robot off the floor under its own power.
The pitch is generality — one platform that learns new tasks instead of one machine per job. Walden's robots run on large behavior models, a class of AI that lets them acquire skills through practice rather than explicit programming. Humanoid's HMND 01 logged more than 90 percent autonomous pick-and-place success at a Siemens electronics plant in Germany, moving 60 totes an hour. Feather Robotics has priced a humanoid worker at $29,990. As we wrote in October, that number only holds if the generality claim survives contact with a real shift.
If it does, task-specific automation starts to look like a dead end: a fleet of single-purpose machines, each carrying its own maintenance schedule, its own spare parts, and its own patch of floor.
That bet has pulled record capital and a $29,990 entry price.
The counter-bet is deployment: fewer capabilities, but uptime you can invoice.
The deployability camp
Ultra's OP1 does four things: packing, sorting, kitting, and picking up items that fall on the floor. It has no legs, no balance controller, and no charging dock — it is bolted in place and draws power from the wall.
The spec sheet is deliberately boring. A 5-by-5-foot footprint, expandable to a 10-by-10-foot work area. Ten pounds of payload per arm. Locking caster wheels so a crew can reposition the unit in an afternoon. Ultra installs it in hours and charges an upfront integration fee plus a monthly rate for hardware and software support. The operator carries the robot as an operating cost, and Ultra carries the risk that it sits idle.
What the machine does not do is decide how to move. Ultra leans on Physical Intelligence, the AI research firm valued at $5.6 billion, for the models that let each OP1 adapt to a specific warehouse layout. The pairing was part of the October 9 announcement, and it is the part that matters most. Ultra is not building the brain. It is buying one and bolting it to a body with no legs.
The evidence Ultra points to is operational. More than 500,000 orders packed. Deployments across New York, Georgia, New Jersey, and Texas. A service model that keeps the robot on the customer's floor rather than in a lab.
Maven Robotics, which emerged from stealth in September with $100 million, makes the same argument from a different angle. Its wheeled dual-arm robots have logged 99 percent uptime or better across 16-hour shifts at a Fortune 250 consumer-goods customer. That figure is measured across production shifts, not across a demo reel. "There's a huge gap between a robot that demos well and one that survives three production shifts a day, seven days a week," said Jack Pearson of RoboStrategy, the fund that led Maven's round.
Series A plus seed, announced Oct 9
$50M Series A led by Framework Ventures; $12M seed from Y Combinator and NextView. · Fortune, 2026
Where the numbers point
Both camps answer the same question — what does a robot have to do before a warehouse operator signs? The humanoid camp answers with capability: the range of tasks one platform can absorb. The deployability camp answers with uptime: the share of a shift a machine is actually working.
| Dimension | Humanoid-first | Deployability-first |
|---|---|---|
| Form | Legged generalist | Stationary or wheeled, task-specific |
| Core bet | One platform, many tasks | Few tasks, high uptime |
| Capital signal | Humanoid $152M; Walden $300M; Figure $1.7B | Ultra $62M; Maven $100M |
| Evidence offered | Demos, prototypes, price points | Production shifts, 99%+ uptime |
| Failure mode | Cost per productive hour; safety certification | Limited task range; re-integration per job |
Comparison drawn from company announcements and Fortune, SiliconANGLE reporting, 2026.
The asymmetry is in the timelines. A humanoid that matches human dexterity could absorb dozens of tasks at once — that is the prize, and it remains ahead of the field. A stationary arm that packs half a million orders is already on the floor, already invoiced, and already generating the operational data that trains the next model.
The economics of uptime
Ultra's pricing model is where the two camps diverge most. A warehouse does not buy the OP1 outright. It pays an integration fee, then a monthly rate that bundles the hardware, the software, and ongoing support. The operator's exposure is a line item, not a capex decision that has to survive a budget committee.
That structure pushes the risk back onto Ultra. If the robot sits idle, the monthly fee still has to cover the hardware that was shipped. The incentive is to keep units running, which is why the company leads with install time and uptime rather than with a dexterity demo.
Humanoid vendors face the opposite pressure. Their capital is priced against a future in which a single platform replaces several kinds of automation, so the near-term cost per productive hour is less urgent to investors than the size of the eventual market. Walden's $300 million and Figure's $1.7 billion are not bets on this quarter's utilization. They are bets on a platform shift.
The labor math sits underneath both. Industrial labor is expensive and getting scarcer in the markets these companies sell into, and mixed-case palletizing — pulling boxes of different sizes and weights into a store-ready pallet — is exactly the kind of repetitive, high-variability work that resists fixed automation. Maven's eight robots work multiple shifts a day at a Fortune 250 customer. Ultra's units run across four states. Those are small fleets. They are also the only fleets with published shift-level numbers.
The bet that does not need legs
Physical Intelligence sharpens the contrast. The firm is building the generalist brain the humanoid camp is counting on — the same class of model that would let a bipedal platform pick up new tasks by watching. Ultra is buying that brain and wrapping it around a machine with no legs, no balance problem, and a power cord.
If the model keeps improving, the body it controls turns into a procurement decision. Generality can arrive in the software while the hardware stays dumb, stationary, and cheap. That is the outcome the humanoid thesis has to argue against, and it is harder to argue as the models get better.
For the operators signing contracts, the question is narrower than the debate. A humanoid worker at $29,990 still has to prove it can hold a full shift, week after week, without a technician standing nearby. A robot that packs 500,000 orders still has to prove it can learn a second job. Neither side has settled it. Ultra's $62 million just moved one of those bets onto more floors.