Uptime Is the Pitch Now — Robot Reality Check

News window: October 12–19, 2026

Editor’s Note

Agility Robotics’ chief business officer said the quiet part out loud this week: for Digit 5, uptime is “the business case.” The pitch is hours of productive work per day, and that’s a very different pitch from dexterity or a general-purpose demo. A new function might attract investors, but a robotics company using operational terms like uptime suggests a real move into practical applications. The same week, the FT ran a piece arguing the AI infrastructure bet is running into “the unforgiving economics of the physical world,” and Microsoft walked away from a UK data center lease over power supply.

Those two stories are the same story from opposite ends. Software margins don’t survive contact with batteries, power grids, and construction crews, and the companies selling physical AI are starting to pitch on the KPIs that operations people actually run their plants by. That’s progress, potentially. It also means the numbers can now be checked, so this week we check them.

Stress Test

Agility Robotics — Digit 5’s 20-hour day

Claim

Agility says Digit 5’s new Nyobolt battery delivers 90 minutes of runtime on a 9-minute charge, a 10:1 run-to-charge ratio against Digit 4’s 2:1, and “more than 20 hours of productive work in a 24-hour day.” Agility’s cofounder also published a first-person account of the redesign in IEEE Spectrum, framed around safe, economical work alongside people.

Demonstrated vs. marketing

The math sort of holds on paper, with caveats (see Spec Check below). But every figure is Agility’s own, and the release is written in the future tense: the battery “will deliver.” There’s no independent test of the charge rate, no data on how the pack holds up over hundreds of fast-charge cycles, and no customer-reported uptime. The IEEE Spectrum piece is a cofounder essay, not an outside assessment.

Deployment reality

Leading with uptime is the right instinct for a company trying to win actual executing customers, and it is (on the surface) a much more practical pitch than another capability demo. The catch is what “productive hours” leaves out. Battery availability is only one input to uptime. Faults, recoveries, and the time someone spends un-sticking a robot all come out of the same 24 hours, and none of that is in this number. Automotive plants, one of the most controlled automation environments there is, still run somewhere around 85 to 95 percent uptime depending on where you measure, and that calculation also carries a lot of assumptions: the entire factory might be “up” while a single part of the line is down, depending on buffer design and so on, so uptime is really only valuable when you know the segment being measured. Agility’s own claim works out to about 83 percent, assuming planned recharging downtime is the only kind you have. It’s also worth asking the question every humanoid pitch should face: for the specific task a buyer has in mind, would a purpose-built system post a better number? Usually, especially in the early days, the answer is still yes.

VerdictPass

This claim of high uptime sounds like a common marketing spin to me. Robot OEMs don’t want to have downtime pinned on their machines if it could be explained away (e.g. other operational issues at the factory that aren’t strictly carried on the OEM’s paper). But customers don’t care where the downtime comes from within the system, at least not directly. I’d be more interested to see what the actual operational time is for one of these machines in an industrial setting with other complexities around the Agility bot. If that uptime is still 20 hours in a 24 hour day, then I might get excited. Actually no, then I would get suspicious.

Promise Robotics × D.R. Horton — “Factory-as-a-Service” for homebuilding

Claim

Toronto-based Promise Robotics signed a multi-year agreement under which D.R. Horton, the largest U.S. homebuilder by volume, will use Promise’s robotic prefab platform for more than 1,250 homes, mostly single-story homes around Phoenix. Promise is opening its third factory there. The Robot Report and HousingWire both covered it.

Demonstrated vs. marketing

This one has the thing most robotics announcements don’t: a named, very large buyer committing to a specific volume. Financial terms weren’t disclosed, and there’s no public data on cost per home or build time against stick-built. HousingWire framed it honestly, as a test of whether industrialized construction can hit the cost and speed targets homebuilders need, not as proof it already does.

Deployment reality

The structure is what makes this credible. The robots work inside a factory, which is a controlled environment, and ship panels to the jobsite, so they avoid the unstructured-site downtime problem that sinks most construction robotics pitches. And the buyer is a homebuilder adopting automation for a production problem it already has, not a tech company looking for a problem to automate. The “as-a-service” model moves the capex risk onto Promise, which is exactly why a relatively low-margin or small builder might say yes. The open question is margin: someone is carrying the cost of three factories, and 1,250 homes across a multi-year deal is still a pilot at D.R. Horton’s scale.

VerdictBuy

Prefab and offsite construction might not sound like the future of construction a Silicon Valley investor would get hot and bothered by, but they are the real next frontier. I like this setup, and the factory-as-a-service model, while buzzwordy, is probably the right move for this market.

Scorecard

Claim	Status	Revisit
Fast-charge humanoid uptime (10:1 run-to-charge) as a category of solution · Agility Robotics / Nyobolt	Online	~Q2 2027
Robotic offsite prefab sold as a service to production homebuilders · Promise Robotics / D.R. Horton	Online	~Q3 2027

Spec Check

Digit 5: “more than 20 hours of productive work in a 24-hour day.”

Take Agility’s own figures at face value: 90 minutes of work, then 9 minutes on the charger. That’s a 99-minute cycle, about 14.5 cycles a day, which works out to roughly 21.8 hours of runtime. So the 20-hour claim leaves about 1.8 hours of slack for walking to and from the charger, handoffs between tasks, and everything else.

The comparison worth making is with Digit 4’s 2:1 ratio, which caps out around 16 hours a day under the same assumptions, minus the walking time and other confounding factors. A jump from 16 to 20-plus hours is interesting if the battery performs as described.

What the number doesn’t include is any downtime that isn’t charging. Twenty hours out of 24 is about 83 percent. And that’s the ceiling. Treat that uptime number as what it is: only a small part of the complex picture that is operational availability.

Funding vs. Function

The trade press and the business press told opposite stories about capital this week.

On the trade side, money kept flowing into physical capacity. Japan offered GlobalFoundries up to $21B for two New York fabs, and ENR reports permitting is done and site work has started. Applied Materials opened a $5B R&D center in Sunnyvale. Tenstorrent raised $1.5B at a $6.2B valuation. And AMT reported August manufacturing-technology orders were the third-highest month on record.

On the general-press side, the tone turned. The NYT, WSJ, and The Economist all ran pieces on stress in AI-related debt, and The Economist asked whether chips are headed for a crash.

These aren’t contradictory. Rounds in the billions, backed by permits, poured concrete, and a government counterparty, are the kind of capital physical AI actually requires. These numbers are only shocking if you ignore the assets these companies, once mature, will have on their balance sheets. The debt stress is concentrated in leased compute and the companies financing it, which is a different bet, but not an independent one. The fabs depend on the industry’s compute needs and trail them, so if the compute buildout stalls, fab demand follows a step behind. And a credit pullback doesn’t distinguish between the two.

Adjacent Industry Check

The week’s construction news explains why Promise Robotics’ pitch is so well timed and well targeted. The 2026 AGC/NCCER workforce survey found contractors still can’t fill craft positions even in a softer market, with data center projects pulling labor away from everything else. Harvard research found immigrants accounted for all net growth in the trades from 2013 to 2024, and that pipeline is shrinking. Construction input prices rose 11.2 percent year over year in September.

Manufacturing solved a version of this problem decades ago by moving work into controlled environments where automation performs reliably. That’s exactly what offsite prefab does for construction, with the added benefit that the labor can now be employed in more advantageous locations (places with infrastructure) rather than the middle of nowhere, as many megaprojects tend to be. The lesson that carries over: the automation that works in construction will mostly be the kind that pulls work off the jobsite. Jobsite robots face far less structure than any factory, and their downtime will be worse than the factory benchmarks buyers hear quoted.

Quick Hits

ABB added an “AI Robot Trainer” to its OmniCore EyeMotion vision system, covering model training through deployment in one loop. No performance figures.
KUKA launched iiQKA.AI Vision, drag-and-drop 2D/3D vision running at the edge on an NVIDIA Jetson Orin module inside the controller.
The FT says Europe’s industrial majors are splitting: Schneider and Siemens toward software, ABB toward physical equipment.
ABB’s new IRB 1700 and 1700ID claim up to 20 percent shorter cycle times and 0.03 mm path repeatability (vendor-reported).
NEURA announced Neuron AI, a multimodal robot foundation model running across five of its embodiments. No independent benchmarks yet.
OnRobot’s 2FGP20 Fork gripper handles open and flimsy cartons up to 16 kg with no compressed air, aimed at replacing custom pneumatic tooling.
Cyngn’s DriveMod EasyLaunch automates a single indoor material route on subscription, with no upfront capital.
Sam Altman confirmed OpenAI is building a humanoid and promised “a cool demo in 2027.” A demo, not a product.
NeodraDynamics showed Neo1, a 4-meter, 1.2-ton hydraulic-electric robot, at CIIF. Trade-show showcase, no deployment data.
DJI Dock 4 swaps drone batteries automatically in a claimed 25 seconds.
New York’s Office for the Aging signed a three-year, $2.1M expansion of Intuition Robotics’ ElliQ companion robot, a public buyer renewing rather than piloting.
Data centers are becoming military targets, and the FT argues designs may need blast-resistant concrete and counter-drone defenses.
Amazon and AEP will explore up to 960MW of X-energy small modular reactors in Indiana for data center power.
Israel will cap new data centers at 200MW, per a report.
Defense-linked launches clustered around AUSA 2026: FORT Robotics’ connectivity layer, XTEND’s XOS platform, and Aslan Motion’s NDAA-compliant drone electronics. All three have reported defense-sector ties.
Perseverance set a Mars-rover distance record of 28.10 miles, credited largely to onboard autonomous navigation.