Ultra raises $50M to lease packing robots, with Physical Intelligence as the brain

Ultra, a Brooklyn-based startup that leases packing robots to warehouses, has raised a $50 million Series A led by Framework Ventures, with Y Combinator participating, Fortune reported on October 9, 2026. Together with a $12 million seed round led by Y Combinator and NextView, the company has raised $62 million. It did not disclose a valuation.

The round comes with a deeper partnership with Physical Intelligence, the robot-model company. Ultra designs, builds and installs the robots; Physical Intelligence provides the AI that lets them learn and improve for each customer’s setup. Chief executive and co-founder Jon Miller Schwartz calls the split “body and brains”.

How does Ultra make money?

Ultra rents rather than sells. Customers pay an integration fee and then a monthly fee for the robot, its software and support, which keeps a large capital purchase off the warehouse’s books. Schwartz told Fortune that revenue is significant and that Ultra has raised prices as demand grew, but he did not give a revenue figure or a fleet size. The one operating number is the company’s count of more than 500,000 packed orders.

Packing is a narrow, repetitive job, but item shapes and packaging change from customer to customer. That is the kind of variation a learned policy is supposed to absorb without a robotics engineer rewriting code for every site, and it is the case Ultra is making for using Physical Intelligence’s models rather than hand-tuned automation.

Why outsource the brain?

Physical Intelligence, founded by researchers from Google DeepMind and valued at $5.6 billion according to Fortune, supplies its software to other robot makers as well as Ultra. For a hardware company with $62 million in the bank, licensing the model means capital goes to robots, installation and service rather than to training runs. We compare the main model suppliers in our foundation-model guide.

The trade-off is dependency. If the model is what makes the robot useful, a competitor that licenses the same model onto similar hardware competes with Ultra on price and service alone. Ultra’s defence has to be what sits around the model: integration speed, uptime and the customer relationships behind a monthly contract. The split also shows up across the stack, as more companies try to own the layer between hardware and applications, which we mapped in the robot OS race.

Schwartz was cautious about humanoids. He told Fortune that humanoid robots reaching a scale similar to Ultra’s are about five years away. The robots producing revenue for Ultra today are purpose-built machines for one task.