Ultra Raises $62 Million to Prove Warehouse Robots, Not Humanoids, Pay the Bills

While venture capital keeps flooding into humanoid robot startups, a Brooklyn company building machines that never stand up just closed one of the year’s most interesting robotics rounds. On October 9, 2026, Ultra announced it had raised $62 million across two rounds, according to reporting by Fortune, and used the moment to deepen its partnership with Physical Intelligence, the AI foundation-model company supplying its robots’ brains.

The money is the headline, but the business model is the story. Ultra does not sell its robots. It leases them under a recurring monthly arrangement, an approach sometimes called robots as a service, or RaaS, and CEO Jon Miller Schwartz told Fortune the model has already let the company raise its prices. Ultra says its machines have packed more than half a million orders. Revenue itself was not disclosed.

The raise, in plain numbers

The funding breaks down into a $50 million Series A led by Framework Ventures, with Y Combinator participating, plus an earlier $12 million seed round led by Y Combinator and NextView Ventures. That is the consistent figure across Fortune, Humanoids Daily, AI Weekly, and Seedtable, all summarizing the same company announcement.

The round lands in what has been a record year for robotics investment. Maven Robotics announced a $100 million Series A in September 2026 to automate mixed palletizing in warehouses, and specialist coverage this year has put total 2026 robotics funding at roughly $18.8 billion, a figure that says less about any single startup than about where capital believes physical AI is headed. Ultra’s raise is modest next to the valuations orbiting humanoid companies, which is arguably the point: the bet is that unglamorous machines doing real warehouse work are a safer place to put money than bipedal demos.

A stationary robot in a standing-up industry

Ultra’s robot, called Operator or OP1, is designed to be uninteresting to look at. It works from a stationary position on locking caster wheels that let it be rolled between work areas, occupies a five-by-five-foot footprint, lifts up to 10 pounds per arm, and plugs into a standard 120-volt outlet. Its listed workflows are packing, sorting, and kitting, the pick-and-place chores that fill warehouse stations.

The design reflects a deliberate tradeoff. A robot that does not walk between stations gives up mobility and gains reliability, because walking is still one of the hardest things a machine can do. Ultra says it has deployments in New York, Georgia, New Jersey, and Texas, and the company focuses its AI on manipulating goods at the workstation rather than navigating the building. Every hour the robot spends packing boxes is an hour it is not falling over, which Schwartz has not been shy about contrasting with humanoid rivals.

The Physical Intelligence connection

The deeper half of the announcement is the tie-up with Physical Intelligence, the San Francisco company founded in 2024 by researchers from Google DeepMind, Stanford, and UC Berkeley to build general-purpose vision-language-action models for robots. Physical Intelligence reported a $5.6 billion Series B valuation in November 2025, led by CapitalG, and specialist coverage in mid-2026 still described it as pre-revenue with no public pricing, structuring partner relationships through individual contracts.

The Ultra relationship is one of Physical Intelligence’s most concrete deployment stories. In February, the AI company shared results from its work with Ultra, including Ultra’s account of a full shift at 96.4 percent autonomy using the π0.6 model, with throughput and reliability improving across successive model generations. Human operators intervene when the model encounters problems, and those interventions double as training data, keeping orders flowing while generating the examples the next model version learns from.

Ultra’s split is clean: it builds the hardware, runs installations, and serves customers, while a specialist AI developer supplies the intelligence. Whether that division of labor scales is the experiment to watch.

This division of labor is worth examining on its own terms. Ultra concentrates on hardware, installation, and customer operations while drawing on someone else’s models, and Physical Intelligence gets experience and real-world data from robots working outside the lab. It is a partnership, but the commercial terms have not been confirmed by Physical Intelligence publicly, and a partner that is still pre-revenue is a partner whose own trajectory could reshape the deal at any time.

Robots as a service, explained

The lease model is Ultra’s answer to the oldest problem in industrial automation: capital expenditure. A warehouse operator considering a robot purchase faces a large upfront cost, an uncertain payback period, and the risk that the technology goes stale before it pays off. Schwartz told Fortune that customers pay an initial integration fee followed by monthly charges for hardware and software support, moving the expense from the balance sheet to an operating line item.

The model cuts both ways. Monthly fees lower the barrier to adoption and align the vendor’s incentives with uptime, since a robot that does not work is a subscription the customer cancels. But RaaS economics live or die on utilization and support costs, including the human interventions that currently keep Ultra’s fleet running. Every intervention is labor that the company pays for or passes along. As the installed fleet grows, the support cost per robot, not just the revenue per robot, determines whether the business works.

It is also worth noting the limits of what has actually been verified. The funding figure is well corroborated, but the claims about revenue growth, pricing power, order volume, and fleet expansion come from Ultra’s CEO speaking to Fortune and have not been independently audited. There are no customer testimonials or third-party revenue confirmations in the public record. That does not make the claims false, but it means the business case rests, for now, on the company’s word.

Why the humanoid jab matters

In the Fortune interview, Schwartz argued that practical non-humanoid robots doing real warehouse work today deliver more immediate real-world impact than humanoids, which he described as prone to falling over and difficult to deploy reliably. That is his opinion, not a settled finding, but it maps onto a genuine industry debate.

The honest state of humanoid deployment in 2026 is pilots, not fleets. Figure AI, Tesla, UBTECH, and Agility Robotics all have humanoids in factory or warehouse pilots, but most run at a fraction of the speed or success rate of a trained human or a specialized robot on the same task. The case for humanoids is flexibility: a machine that can open a door, climb three stairs, and deliver a part to a workstation that moves every six weeks may be the only automation that fits. The case against them is that narrow, proven machines like OP1 ship revenue now.

What to watch next

Two questions will decide whether this round ages well. First, can Ultra demonstrate the unit economics of RaaS at scale, publishing or proving that monthly revenue outruns hardware depreciation, support labor, and intervention costs as the fleet grows? Second, does the Physical Intelligence partnership mature into a durable arrangement, or must Ultra eventually own its intelligence stack? For the industry, Ultra tests a proposition: the fastest path to useful robots runs through boring machines, deployed one workstation at a time.