Everyone wants to talk about the robot’s brain. I want to talk about its battery.
On October 15, Agility Robotics announced that Nyobolt is the battery technology partner for Digit 5, its humanoid. The numbers in that release are the most important robot spec I read all week:
- 90 minutes of runtime
- 9 minutes to charge
- A 10:1 run-to-charge ratio, up from 2:1 on Digit 4
- More than 20 hours of work in a 24-hour day
Agility’s Chief Business Officer, Daniel Diez, put it plainly: “For our customers, Digit’s uptime is the business case.”
He’s right. And it lines up exactly with the last post , where Anthropic’s research showed that robots can do most physical tasks but are rarely cheaper than people. The cheapest way to make a robot cheaper is to make it work more hours.
Duty cycle, the oldest spec in the book
In electrical engineering, duty cycle is the fraction of time something is on. Motors, power supplies, welders, test fixtures. Every one of them has a rated duty cycle, and if you ignore it you either cook the part or you buy three of them when you needed one.
For a battery robot, duty cycle is run time over run time plus charge time.
Digit 4 at 2:1 2 / 3 = 66.7% about 16 hours a day
Digit 5 at 10:1 90 / 99 = 90.9% about 21.8 hours a day
That’s an idealized ceiling. It assumes no faults, no maintenance, no shift changes and no waiting on work. Agility’s own claim is “more than 20 hours,” which fits under that ceiling.
Here’s a day at 10:1, to make it concrete:
06:00 shift starts, battery full
07:30 dock, charge 9 minutes
07:39 back to work
09:09 dock, charge 9 minutes
09:18 back to work
...
15:30 a forklift honks twice
...
05:51 dock, charge 9 minutes
About 14 or 15 charge stops in 24 hours, each shorter than a coffee break.
Why this is an ROI story
Anthropic’s report on what work robots can do prices robots the way a finance team would. Fixed costs are spread over a roughly 10-year service life at an 8% cost of capital. Operating costs like maintenance, part-time supervision and energy sit on top.
Fixed cost doesn’t care how many hours the robot works. So the more hours you get, the cheaper each hour gets.
Going from 16 hours to about 21.8 hours is about 36% more working time from the same machine. Spread the same fixed cost over 36% more hours and the fixed cost per productive hour drops by roughly a quarter. That’s my back-of-the-envelope math, not Agility’s, and real downtime will eat some of it. But that’s the shape of the curve, and it’s why Agility led with uptime instead of intelligence.
There’s a catch the Anthropic report points out. Cleaning robots “might work longer hours,” but can be up to 10 times slower than people. Hours only help if each hour does real work. Duty cycle multiplies throughput. It doesn’t replace it.
The questions I’d ask as a test engineer
A press release is a datasheet with the hard rows left blank. Here are the rows I’d want filled in before I’d put this in a business case.
What’s the charge window? “Charges in 9 minutes” from what to what? Fast charging usually covers part of the range. The release doesn’t say, and I’d ask.
What’s the cycle life? At about 14 charges a day, a robot runs through roughly 5,300 charge cycles a year. That’s my math from Agility’s numbers. Nyobolt says its technology combines fast charging with “exceptional cycle life,” but there’s no number in the release. That number decides when you’re buying a new pack.
What happens in heat? Fast charging makes heat. Warehouses and factories in summer make more. I’d want to see charge time and capacity across temperature, not at room temperature on a bench.
How does capacity fade? Ninety minutes on day one is a spec. Ninety minutes in year three is a business case.
When can I actually get it? Agility says early access to Digit 5 with this battery is expected in the first half of 2027, with more availability over the following six to nine months. So today it’s a forward-looking claim. Agility’s release says so itself, at length.
None of that is a knock on Agility. It’s what anyone should ask about any battery-powered machine they plan to run around the clock. I’d ask the same about a fleet of handheld scanners.
Uptime is the business case
The software world learned this a long time ago. We stopped asking “does the service work?” and started asking “what percent of the time does it work, and what does an hour of downtime cost?” That’s an SLO.
Robots are getting the same treatment, and the battery is the first SLO they hit. A humanoid that can do a task but sits on a charger a third of the day is a part-time employee with full-time capital costs.
Ten to one doesn’t make Digit smarter. It makes Digit cheaper per hour. Right now that’s worth more.