Robotics Funding: Ultra's $62M Warehouse Bet Amid Humanoid Boom

A busy October for robot money

Robotics funding in October 2026 is moving on two tracks. One is a wave of very large humanoid rounds. The other is a set of smaller deals that pay for robots already doing work on customer floors. Brooklyn-based Ultra is the clearest example of the second kind. The warehouse robot maker announced a $50 million Series A led by Framework Ventures, with Y Combinator participating. Added to an earlier $12 million seed round, its disclosed funding now totals $62 million.310 Y Combinator and NextView Ventures co-led the seed.6

Ultra arrived during a run of deals elsewhere in the stack. Robot-data startup Mecka AI raised a $60 million Series B led by Sequoia, with Nvidia and Microsoft's M12 participating. TechCrunch had earlier reported that the company was nearing a round at a $500 million valuation.15 Minerva Humanoids came out of stealth with a $10 million pre-seed round led by General Catalyst. Its semi-autonomous humanoid is built for hazardous energy and public-safety work.14 DeepSeek is reportedly moving to lead a $400 million seed round for Winverse AI, a stealth Chinese startup building AI for robots.24 Sudo AI, which builds robot foundation models, raised more than $149 million equivalent in early October.23

Read together, these deals show where investors see value right now. Money is going to robot bodies, to the models that control them and, more and more, to the data and deployment work that links the two.

Ultra's model: rent the robot, borrow the brain

Ultra's product is the Operator, or OP1. It is a stationary, two-armed robot that handles e-commerce packing, sorting and kitting. It runs off a standard 120-volt outlet and an ethernet line.1 Ultra says it can be installed in hours without building changes.1 Its footprint is about five by five feet, and each arm carries up to 10 pounds.3 Ultra does not sell the machine. Customers pay an integration fee and then a monthly charge for hardware and software support, a robots-as-a-service arrangement.3 CEO Jon Miller Schwartz has said this structure produces significant revenue and has let the company raise prices. He has not given a revenue figure.10

The more important design choice is that Ultra does not build its own AI. The robot's control models come from Physical Intelligence, a foundation-model company.1 In February, the two companies reported that Physical Intelligence's π0.6 model ran an Ultra robot through a full warehouse shift at 96.4% autonomy while packing real customer orders.5 Several outlets point out that this figure comes from the companies themselves and has not been independently verified across Ultra's fleet.59

Ultra says its robots have packed more than 500,000 orders at sites in New York, Georgia, New Jersey and Texas.1 One deployment, at third-party logistics provider Highline Commerce, uses three units to handle up to 30% of that customer's order volume.1

Where the coverage agrees, and where it doesn't

The reporting agrees on the main facts: the round sizes, the lead investor, the service model and the Physical Intelligence link. It also shares the same caution. The 500,000-order figure is a cumulative company claim, not tied to a fleet size, an operating period or per-robot economics.102 One analysis says the funding starts Ultra's real test rather than finishing it.2 That fits the evidence. Leasing shifts capital risk from the customer to the vendor, so Ultra recovers its costs only if machines keep running for months without frequent site visits.2

There are inconsistencies. A few headlines said Ultra "raised $62 million," which merges two separate rounds. The more careful reports separate the new $50 million from the earlier seed.610 One outlet described the OP1 with forklift-style grippers and a 10-foot shelf reach.8 That does not match the stationary, bimanual workstation robot most other reports and Ultra's own specifications describe.13 The workstation version is the better-supported account.

Reports also differ on how much Physical Intelligence has raised. One says more than $400 million.28 Another cites a $600 million round at a $5.6 billion valuation, with talks reportedly underway for $1 billion at an $11 billion valuation.1 A company profile lists $1.1 billion in total funding and an $11 billion target valuation.26 The likely explanation is that some outlets are using older totals. Either way, the model supplier behind Ultra is worth many times more than the hardware company using its models.

The humanoid mega-round context

The split between hardware and AI matters because of the scale of humanoid funding. One tracker counts 25 disclosed equity rounds for pure-play humanoid companies between August 2025 and July 2026, totaling $6.2 billion. The median round was $145 million.12 Tracxn reports $6.61 billion raised across 42 humanoid rounds through September 2026, up from $2.05 billion a year earlier.13 A Dealroom chart puts year-to-date venture investment in humanoid robotics at $11 billion, more than double 2025's full-year record.20 The gap between these figures comes down to definitions. Another analysis found that first-half 2026 physical-AI funding ranged from $18.8 billion to $55.8 billion depending on scope.25 Every measure shows sharp growth. None of them agrees on the actual size.

The leaders are well known. Figure AI tops most rankings, though estimates of its total vary: more than $1.7 billion,11 $1.75 billion,13 or about $1.9 billion.21 Figure has also agreed to a compute deal with Nscale for up to 100,000 Nvidia Vera Rubin GPUs. The initial commitment is $3.5 billion, with intent to scale past $6 billion, to train its Helix model.21 NEURA Robotics announced a Series C of up to $1.4 billion. Apptronik's Series A extension of $520 million pushed that round past $935 million.16 Skild AI raised $1.4 billion in January for a robot-agnostic "brain," which reportedly took its valuation above $14 billion.16 Chinese companies including Galbot, AI2 Robotics, Sharpa and Galaxea have each raised hundreds of millions.11

Compared with these numbers, Ultra's $50 million round is modest. One market-watch note called it an incremental raise. The same note argued it could matter more than its size suggests, because it tests whether embodied foundation models work on a commercial fleet.28

Foundation models may be turning into a commodity

That argument draws support from model releases this month. Nature Will, a little-known lab, announced Natura-1. It is a 50-billion-parameter mixture-of-experts model for dexterous hand control, pre-trained on more than a million hours of first-person human video.21 Forbes commentary read the release as a sign that the cost of building robot foundation models is falling fast. On that reading, value shifts to post-training: making one robot useful for one job at one company.21 Odyssey's world model, Odyssey-3, reportedly trained a driving policy on 20 hours of generated video that then drove on real streets in India. Independent researchers have not yet reproduced that result.27

If that reading is correct, Ultra's position makes sense. Building a model may stop being the defensible part of the business. Field data, installation routines and customer relationships could become harder to copy. Ultra gets models it could not build alone. Physical Intelligence gets messy real-world data that lab benchmarks cannot provide.23

Warehouses favor narrow, practical robots

Other warehouse deals point the same way. Destro AI, founded in 2025, raised an $8 million seed round co-led by Base10 Partners and Bonfire Ventures. It sells a hardware-agnostic coordination layer that decides which robot does which task.4 Its founder argues that hardware is no longer the bottleneck and that shared intelligence is.7 Ultra competes with better-funded players such as Dexterity, valued at $1.8 billion. The field also has a warning in Covariant. After raising $245 million, Covariant was effectively absorbed by Amazon in 2024 through a hire of its founding team.1

The main conclusion from this coverage is that humanoids are drawing most of the capital, but most of the proof is coming from narrower machines doing specific jobs. Ultra's next numbers will be the test: fleet size, how often humans have to step in and whether customers renew. They will show whether splitting the robot's body from its brain can scale, or whether each new warehouse becomes a custom engineering project.2 Until a humanoid company publishes deployment data of comparable detail, a $50 million round for robots that pack boxes says more about where robotics actually stands than the billion-dollar rounds do.