Apptronik’s Apollo 2 in operation. Today Apptronik announced Apollo 3 and a new data factory, Robot Park.
Apptronik
“I would argue the industry today is essentially all prototypes,” Apptronik CEO Jeff Cardenas told me yesterday while announcing Robot Park, a massive physical AI data factory, Apollo 2 – which has existed for some time – and teasing Apollo 3, the next-generation humanoid robot Apptronik will unveil next year. Apollo 3 will be the actual shipping robot that Apptronik is counting on to do real work at high speed and quality, in exchange for cold hard cash.
In robotics we see a lot of demo reels and highlight videos: robots folding laundry, sorting parts, pouring a drink, or making eggs. That’s exciting, and the rate of improvement from quarter to quarter is impressive, but the reality today is that almost every humanoid released to date is essentially still a prototype: they can’t do what a human can at required levels of speed and accuracy.
And while that’s still true of Apptronik’s current Apollo 2 robot, it won’t be for Apollo 3.
“Apollo 3 will be a product. It will be a mature, early product, but still a product,” Cardenas told me in a brief call yesterday. “You can think of Apollo 2 as really a prototype for scaled pilots and data collection, and the next version of Apollo will really be the product that will scale to the market.”
A data factory to match the robot factory
That scale will be depending on hardware, to be sure. But it will also be dependent on software: the physical AI that powers humanoids. That’s why the Austin-based company has built Robot Park, a 90,000-square-foot facility it describes as a flagship data-collection and training site for humanoid robots.
According to Cardenas, it all started with a dream and some beer.
“The Robot Park is really the embodiment of a dream we’ve had for a long time,” he said. “When we first started, I would go get beers with my co-founder, Nick, and dream up this idea of robots building robots on the Colorado River. I called it Willy Wonka’s Robot Factory.”
The version that exists today is less whimsical and more industrial, but the logic is the same. Apptronik already runs a factory that builds robots. Robot Park is meant to be the other half of the flywheel.
“Just like we have a factory building robots, we also have a data factory as well,” he said.
Apollo will ship in both wheeled and bipedal forms
Apptronik
Inside the expanded Austin facility, fleets of Apollo 2 robots — in both bipedal and wheeled-base configurations — run through logistics, manufacturing, retail and other customer-driven tasks, generating the real-world data that embodied-AI models need to learn. Through a mix of teleoperation and autonomous execution, the company says, the robots continuously produce high-quality training data. That’s the data that, along with the updated hardware of Apollo 3, will drive humanoid performance to the level of an actual shipping product with hard ROI for customers.
Crucially, Austin is not the only such site.
Apptronik says it has deployed similar data-collection workflows across a growing network of Robot Parks, including at research partner Google DeepMind and at customers like Mercedes-Benz and GXO, the high-volume logistics giant. More locations are planned.
How many robots are running across all of it? Cardenas wouldn’t say. “We haven’t released the specific numbers of the Apollo 2s, but lots of robots,” he said. I’m guessing it’s going to be orders of magnitude larger than the current U.S. champion for robot data factories in Boston, owned by Tutor Intelligence.
The reticence about the number of robots is kind of part of the story. Apollo 2 has existed since February of last year without a big announcement, appearing suddenly on Apptronik’s website with little fanfare. Cardenas announced Apollo 3 in much the same way, offhandedly during our conversation. The point, I guess, is a sort of Apple-like dislike of getting into spec wars with competitors on every little detail, and instead focusing on raw capability.
But we should expect Apollo 3 reasonably soon, likely by next year.
“I don’t know that I would wait another 18 months to launch a robot again because I’ve gotten a lot of questions about Apollo 2 over the last year,” Cardenas said.
Apollo will have both wheels and legs
One of the more practical reveals is that Apollo 2 ships in two flavors: a bipedal robot and a wheeled-base version. That mirrors what I’ve heard repeatedly from people deploying humanoids: industrial and logistics customers keep asking for wheels, for stability, battery life and predictability.
Legs burn battery just standing still. Wheels are more efficient, and you can pack a bigger battery low on a wheeled platform, creating a very stable platform with a low center of gravity.
“I still think that the bipedal humanoid has the highest ceiling in terms of what it can do, so it has the broadest workspace that it can perform inside of,” he said. “But there’s a challenge with bipeds in that they can fall over.”
That matters if you’re going to do deployments in the tens of thousands or millions. “We’re really focused on how to solve safety and how to be smart about the ways we solve safety for bipedal humanoid robots,” he said.
Wheels are a positive in another way too: regulatory compliance. Apollo’s wheeled configuration is designed to conform with existing safety standards for industrial mobile robots, which makes it easier to drop into a customer’s operation today.
The two form factors also feed each other. “
We can get big fleets of wheeled systems out to pave the way for the bipedal systems as they mature and come online,” Cardenas said. “I think in manufacturing and logistics you’re going to see a lot of wheels, especially early on. As you start to look toward future markets, I think that’s when the bipeds will really shine.”
Chief Commercial Officer Barry Phillips put the modularity in customer terms. “For truly useful humanoid robots, safety and reliability have to advance alongside capability,” he said in a statement. “The modular design of Apollo is a direct response to customer demand for adaptable automation.”
The leap to Apollo 3
Everything Apptronik learns from Apollo 2 will flow directly into Apollo 3, the version Cardenas keeps calling “the product.”
“You can think of Apollo 2 as really a prototype for scaled pilots and data collection, and the next version of Apollo will really be the product that will scale to the market,” he said.
The differences, as he describes them, are less about a flashy new trick and more about the unglamorous fundamentals that separate a demo from a deployable machine.
“Apollo 3 will just be much better. It’s much more scalable and much more focused on BOM cost,” he said, referring to bill-of-materials cost, the per-unit economics that decide pricing strategy.
The company has reworked the end effectors and the sensor stack, and Cardenas says its approach to safety will be distinctive: “How we solve safety is going to be unique on Apollo 3, so things like safe perception are really important to us.”
The Google DeepMind question
Apptronik’s data has a destination: Gemini Robotics, Google DeepMind’s foundational AI models for robots. The two have been partners (DeepMind is both investor and collaborator) for nearly two years.
Which raises an obvious question: if Apptronik’s real-world data keeps making DeepMind’s models smarter, does that intelligence eventually flow to Apptronik’s rivals too?
Cardenas didn’t dodge it.
“Google has been honest with us from the very beginning about their ambitions to build the Android for robotics,” he said. “That means their model will go to many different partners.”
His bet is on the depth of the relationship rather than exclusivity. “We’ve been in the trenches for several years building and advancing Gemini,” he said. “We like to think we have a special relationship, and we’ll see how that evolves in the coming months and years ahead.”
It fits a company that is ambitious and, by its founder’s own admission, scarred. “We think we have an opportunity to build one of the biggest companies in the world,” Cardenas said, before adding that years in robotics have left him “humbled by the challenge many times, so we’re pragmatic about how to get there.”
‘The space race of our time’
The last thread of our conversation wasn’t part of the press release: the geopolitics now wrapping itself around humanoids.
Apptronik builds and patents its own actuators — roughly half a robot’s component cost — and Cardenas sees that vertical integration as a national issue, not just a corporate one.
“My view is that modern robotics is the space race of our time,” he said. Whoever wins it “will be a big part of the future, both for national competitiveness and national security.”
Apptronik, he notes, has “built nearly eighty iterations of electric actuators — all sorts of different variants and technologies: rotary, linear, liquid-cooled, quasi-direct drive, series elastic, tendon-driven,” and he argues it has “one of the best actuation teams in the world right here in Austin, Texas.”
Over time, Apptronik will add more vertical integration to its component source, he added, but the company will do it intelligently and appropriately. In other words, he’s not chasing vertical integration at all costs.
For now, the message is consistent with the rest of the day’s news: stop watching the demos, and start watching what the robots do every shift.
“The industry has spent years showing what robots can do in demos,” Cardenas says. “We’re focused on what they can do every day on the job.”