Mapping Apptronik: How Robot Park, Apollo, and Google DeepMind fit together

Apptronik's June 30 announcement is less a product launch than an ecosystem statement. The Austin company unveiled the expansion of Robot Park, a 90,000-square-foot "data factory" for humanoids, showed off Apollo 2 in bipedal and wheeled forms, and teased Apollo 3, the version of its humanoid robot that CEO Jeff Cardenas calls the actual product. 

Taken together, the news maps out where Apptronik sits in the humanoid stack and who it's wired into.

I chatted with CEO Jeff Cardenas two days ago:

Here's the ecosystem view of this announcement ...

The AI layer: Google DeepMind and Gemini Robotics

Apptronik's most important connection isn't a customer — it's a partner.

Apptronik's physical AI data, soon to grow enormously thanks to Robot Park (and other Robot Parks around the world), feeds not only its own Apollo humanoids, but also Gemini Robotics, Google DeepMind's foundational AI models for robots.

Google DeepMind is both an investor and a collaborator, a relationship running close to two years.

This is a super-interesting business model, because by virtue of this relationship, Apptronik will inevitably be helping to train a physical AI model that its competitors could eventually use.

The link is non-exclusive by design, and Cardenas is very clear-eyed about it. 

"Google has been honest with us from the very beginning about their ambitions to build the Android for robotics," he told me. "That means their model will go to many different partners."

But Cardenas' hedge against commoditization is depth rather than lock-in: 

"We've been in the trenches for several years building and advancing Gemini ... we like to think we have a special relationship." 

On the map, this makes Apptronik one spoke on a DeepMind hub that will eventually include rivals. That's a business risk, in a sense. But Cardenas is betting that being earliest and deepest is worth more than being only. And being Google's key partner here is a huge benefit too: Google's "Android for robotics" is being trained largely on Apptronik robots. That will mean that there will be a very tight connection between Apptroniks' bodies and Google's brains ... possibly tighter than any other robot manufacturer coming along later can develop.

The data layer: Robot Park and its peers

If Gemini is the brain, Robot Park is the feedstock. 

Fleets of Apollo 2 robots run logistics, manufacturing and retail tasks across the Austin facility and a growing network of sites, generating training data through a mix of teleoperation and autonomous execution.

"Just like we have a factory building robots, we also have a data factory as well," Cardenas said.

That puts Apptronik in an emerging category of humanoid data-collection operations, the same competitive lane as Tutor Intelligence's Boston facility, which was probably formerly the largest such data factory in the U.S. 

Apptronik won't give robot counts: Cardenas told me "we haven't released the specific numbers of the Apollo 2s, but lots of robots," but at 90,000 square feet, Robot Park is bound to be bigger than Tutor Intelligence's facility (which I've been inside).

The customer layer

Apptronik's data doesn't just come from a lab sandbox; it comes from multiple global deployments.

The company says Robot Park-style workflows are running at research partner Google DeepMind and at customers including Mercedes-Benz and logistics giant GXO. Those relationships do double duty: revenue-generating pilots and real-world data sources feeding the model that improves the next robot.

The hardware layer: two form factors, one platform

Apollo 2 currently ships as both a biped and a wheeled base.

This is a deliberate hedge across the ecosystem's near- and long-term needs. Wheels solve today's industrial demand (stability, battery life, and compliance with existing safety standards for industrial mobile robots); legs keep the long-term ceiling high for additional uses cases (stepping from loading docks to trailers, walking up stairs from the factory floor to the office, and maybe even the home).

"I still think that the bipedal humanoid has the highest ceiling in terms of what it can do... But there's a challenge with bipeds in that they can fall over," Cardenas said. The strategy is to let one seed the other: "We can get big fleets of wheeled systems out to pave the way for the bipedal systems as they mature and come online."

Apollo 3 is where this converges into a commercial product. 

"Apollo 3 will be a product. It will be a mature, early product, but still a product," Cardenas said — sharpened on BOM cost, end effectors, sensors and what he calls "safe perception." 

Cardenas didn't give me a ship date: the only timing signal is his reluctance to "wait another 18 months to launch a robot again."

The supply-chain layer: actuators and geopolitics

The node most relevant to ecosystem-watchers is vertical integration. 

Apptronik builds and patents its own actuators — roughly half a robot's component cost — and Cardenas frames this and other parts of vertical integration as essential national infrastructure.

"My view is that modern robotics is the space race of our time," he said. Whoever wins "will be a big part of the future, both for national competitiveness and national security." 

The company has "built nearly eighty iterations of electric actuators," he noted, and he argues its actuation team is "one of the best... in the world right here in Austin, Texas." 

Expect more here: he promised further supply-chain moves later this year. But he's going to be pragmatic about it: vertical integration where it makes sense, sourcing where that would be prohibitive in cost or time.

Where Apptronik lands on the map

Put all the layers together and the position is distinctive: a hardware maker deeply coupled to a global-scale frontier AI lab, running a distributed data-collection network fed by real customers, while pulling critical components in-house. 

This is impressive, and it's a differentiated position compared to 1X, which makes Neo, and Figure, which makes Figure 03.

Which model is best and who will ultimately win, of course, remains to be seen.