Ultra raises $62 million to rent warehouse robots that run Physical Intelligence software

A warehouse that rents from Ultra pays an up-front integration fee to have the robots installed, then a monthly fee for hardware and software support ^[3]. Jon Miller Schwartz, Ultra's chief executive and co-founder ^[14], told Fortune's Term Sheet newsletter that this structure is why Ultra has booked significant revenue, since customers do not have to lay out large amounts of capital ^[4]. He did not say how much revenue, and neither report includes the monthly fee, Ultra's valuation, the terms of the Physical Intelligence deal or who owns the robots at customer sites.

Ownership decides how much cash the model consumes. If Ultra keeps title to the machines, each new site is money spent up front and recovered month by month. In that case the $50 million Series A, about 81% of the total ^[16], is partly fleet capital. It arrived roughly two years after Ultra's founding in 2024 ^[11]. Short installs help a lessor. Ultra says its robots go in within a matter of hours ^[13], and a company that installs in hours spends less on each new site before the monthly fees start.

Physical Intelligence has raised far more money than its hardware partner. Its November round alone was about 9.7 times everything Ultra has raised ^[17]. Crypto Briefing reported that Pi's pi0.6 model reached 96.4% autonomy during full shifts in real-world warehouse settings, citing deployment data released in February 2026 ^[18]. If that rate carried over to Ultra's fleet, the robots would still need human help for 3.6% of each shift ^[20].

Schwartz says the model has gained enough traction that Ultra has been able to raise prices ^[5]. I think that claim is the strongest support for the leasing model, and the hardest to credit to Ultra alone. A customer paying more each month could be paying for Ultra's hardware and installation work, or for gains in Pi's software that reach the robots through the rental. Fortune's case for the split is that Ultra sticks to building robots while Pi gets out-of-the-lab data to train its models ^[9]. That data gives Ultra something to bargain with if Pi reprices its software. The view fails if Ultra's price increases stop once rival robots running Pi's software ^[10] reach the same third-party logistics sites where Ultra's team does its installs ^[15].