Ultra Robotics $62M Raise Backs Warehouse Robots-as-a-Service

A Series A built on packed boxes, not demos

Brooklyn warehouse-robotics startup Ultra has announced $62 million in total funding. The figure combines a $50 million Series A led by Framework Ventures, with Y Combinator participating, and an earlier $12 million seed round led by Y Combinator and NextView.110 Alongside the money, the company said it is expanding its partnership with Physical Intelligence, the robot-AI developer that supplies the models running Ultra's machines.14

Most coverage leads with the funding total, but the business behind it is more interesting. Ultra does not sell robots. It rents them to warehouses under a "robots as a service" arrangement: customers pay an upfront integration fee to get units installed, then a monthly fee that covers hardware and software support.114 CEO and co-founder Jon Miller Schwartz said demand has been strong enough for the company to raise prices.110 He also said Ultra has booked significant revenue, but he did not give a number.17

It is a Series A shaped by the current robotics market. Investors are not paying for a humanoid prototype. They are paying for a narrow, stationary machine that is already doing paid work on real warehouse floors.

What Ultra actually deploys

Ultra's main product is the Operator, or OP1, a stationary dual-arm robot that packs orders for shipment.13 One detailed account describes a footprint of about five by five feet and up to ten feet of vertical reach.12 The unit sits on locking casters, so workers can roll it into place, and it runs on a standard power outlet and an Ethernet connection.317 Ultra markets it for packing, sorting and kitting, and says a single robot can handle up to 1,000 items a day once running, with setup taking only a few hours.317

The company says its robots have packed more than 500,000 orders at U.S. sites.15 Several outlets add that deployments span New York, Georgia, New Jersey and Texas.13 At Highline Commerce, a third-party logistics provider in Brooklyn, three Operator units reportedly handle up to 30% of order volume.13 These are the company's own figures, passed along by the press rather than independently audited, and at least one outlet says so plainly.

The team behind the hardware also helps explain the bet. Schwartz trained as a mechanical engineer and spends time installing robots at logistics sites himself.9 He has founded companies before, including the 3D-printing manufacturer Voodoo Manufacturing, which went through Y Combinator. Ultra was founded in Brooklyn in 2024 and joined YC's Summer 2024 batch.3

The body-and-brains split

The Physical Intelligence partnership is the strategic core of the story, and coverage agrees on how it works. Ultra designs, builds, installs and services the robots. Physical Intelligence supplies the AI that lets each unit adapt to a customer's layout instead of running a fixed script.112 In return, Physical Intelligence gets messy real-world data that is costly to collect in a lab.1018

The relationship is not new. Physical Intelligence published an account of Ultra's warehouse deployments in February, describing how successive versions of its π models improved as Ultra's operating data went into training.3 In one partner-reported example, a model ran an Ultra robot packing customer orders for a full shift at 96.4% autonomy. Human operators stepped in when the model faltered, which produced more training data.413 That figure comes from a single deployment and says nothing about Ultra's whole fleet.4

Physical Intelligence is a much bigger company than its partner. It was founded by a team out of Google DeepMind and is valued at $5.6 billion.1 One outlet reports that the valuation came from a $600 million round led by CapitalG last November, and that the company is reportedly in talks to raise about $1 billion at an $11 billion valuation. Its software already runs robots from several other makers.112

Where the analysis diverges

On the facts, the reporting is consistent. The disagreement is over what the partnership means for Ultra's long-term position.

The optimistic view treats it as a sensible division of labor. Ultra focuses on building and deploying hardware, and Physical Intelligence improves through field data.1 Some commentators go further and describe a data flywheel: more robots produce more data, which yields better models, which attracts more customers.2 Another argues that the intelligence layer is consolidating around a few foundation models, leaving startups like Ultra to compete on hardware design and the unglamorous work of integration.

The skeptical view is sharper. RobotAIGeek argues that Ultra licenses its most important capability rather than owning it. On that reading, a warehouse that signs a multi-year contract is effectively taking on the execution risk of two companies.12 The same analysis notes that no financial terms, exclusivity clause or contract length have been disclosed. That leaves open who owns the data generated on customers' floors.12 AI Weekly's editors put it more bluntly: Ultra's pricing power depends on a model it does not own, so the leverage in the relationship sits with Physical Intelligence.13

The skeptics have the stronger case on structure. Physical Intelligence licenses to multiple robot makers, so having its model does not set Ultra apart.12 What can set Ultra apart is how reliably its hardware and service teams turn a shared model into packed orders, and the 500,000-order count is the best public evidence that it can.12 Still, the dependency is a real risk to manage, not a fatal flaw. Building a competitive robotics foundation model in-house would cost far more than $62 million.

One report stands apart from the rest. Hungarian tech site LaVX describes total funding above $100 million, a CEO named Andrew Liu, and robots that do palletizing and machine tending for manufacturers.8 That contradicts the Fortune-based coverage, which consistently names Schwartz as CEO, puts total funding at $62 million and describes packing robots at logistics sites.1511 Readers should treat the LaVX version as unreliable.

Reading the round: valuation, investors and economics

For anyone tracking startup valuations, the main gap is that Ultra has not disclosed one.78 Revenue, unit prices, margins and installed robot counts are all undisclosed too.513 A $50 million Series A following a $12 million seed is a large step up, which suggests investors saw enough repeat demand to fund expansion without a public customer list.7

The choice of lead investor is notable. Framework Ventures is better known for crypto and gaming bets than for industrial automation, while Y Combinator's presence in both rounds signals continued support from Ultra's earliest backer.18 Ultra says the money will go toward engineering, manufacturing, a larger hardware fleet and wider commercial deployment.1117

The robots-as-a-service model is also less novel than some coverage implies. Ambi Robotics used it earlier, raising $26 million in 2021 and $32 million the next year, and signed a four-year sorting-robot subscription deal with OSM Worldwide. Exotec, Amazon's in-house fleet and Travis Kalanick's logistics venture Atoms also compete for warehouse automation spending.116

The economics are the main open question. Because the vendor carries the hardware cost while customers pay over time, subscription robotics is capital-intensive. Margins depend on how much each unit handles and how little human help it needs, and Ultra has not published either figure.7 Hardware depreciation adds pressure: one legal commentator notes that onboard computing chips age faster than robot arms, which makes how the fleet is financed a strategic decision.15 In practice, this round is partly working capital for carrying hardware until the monthly payments add up.

The bottom line

Ultra's raise is a vote for deployment over spectacle. Schwartz argues that non-humanoid machines are doing more real-world work than the humanoids that get most of the attention, and he estimates humanoids are about five years from operating at scale.17 Investors appear to agree, at least for now.

The hard tests come next. Ultra has to add customers faster than it adds hardware obligations, keep its price increases, and show that improvements in Physical Intelligence's models lead to fewer interventions on the warehouse floor.37 If those things happen, the company is a strong template for how frontier robot AI reaches paying customers. If they don't, its biggest strategic asset will turn out to be its biggest dependency.