OpenAI Builds a Humanoid, Google Licenses the Brain

In early September 2026, OpenAI CEO Sam Altman confirmed on the Sources podcast that OpenAI will build its own humanoid robot — hardware included, not just a model for someone else's chassis. A month earlier, Google DeepMind went the opposite direction, shipping Gemini Robotics 2 as a licensable whole-body intelligence layer that runs on other companies' robots, including the Apptronik Apollo 2 demo.

So the race to put a robot in your living room now has two completely different strategies. Here is what each one actually means for home buyers.

What OpenAI actually announced (and what it did not)

Altman's answer was short. Asked whether OpenAI was developing humanoids or machines for data centers, he said "Both" — adding "We will definitely do a humanoid. We will do other form factors as well."

That is a strategy confirmation, not a product announcement. As of September 2026, OpenAI has disclosed no prototype, no delivery schedule, no production target, and no manufacturing partner. Treat it as intent with hiring behind it, not a shipping product.

The hiring is the real signal. OpenAI has roughly 19 open robotics positions in San Francisco, including four focused on actuators plus a technical program manager for actuator work. Actuator design roles covering motors, transmissions, sensing, and thermal architecture — with manufacturability requirements — go far deeper than what you would need to drop a model onto an existing commercial chassis. Actuators can account for 40 to 60 percent of a humanoid's bill of materials, so building this competence in-house is a multi-year commitment, not a side project.

Context matters too: OpenAI participated in Figure AI's $675 million funding round in 2024, but Figure ended that collaboration in February 2025 to develop its own models in-house. OpenAI's previous robotics project shut down in 2021, reportedly in part over insufficient physical interaction data. Building and operating its own robot fleet would give OpenAI direct access to exactly that data — what the robot attempted, how hard it gripped, what moved as a result.

Altman also set expectations honestly: industrial infrastructure and manufacturing come first, household adoption later. Nobody should expect an OpenAI home robot next year.

What Google actually shipped

Gemini Robotics 2, announced July 30, 2026, is a real product developers can use today — and that is the fundamental difference. It has three parts:

• Gemini Robotics 2 (VLA): a vision-language-action model that converts vision and language input into motor control, capable of whole-body humanoid control "from feet to fingertips" plus dexterous two-handed manipulation.
• Gemini Robotics ER 2: an embodied reasoning model for multi-step task planning and multi-robot teamwork, available now on Google AI Studio.
• Gemini Robotics On-Device 2: a local model that runs on the robot itself and adapts to entirely new robot bodies with only a few hours of data.

The headline demo ran on Apollo 2 — Apptronik's current-generation humanoid, which operates in both bipedal and wheeled-base configurations at its Austin Robot Park and customer sites. Apptronik's data from teleoperation feeds directly into the DeepMind collaboration. That is the licensing model working in public: Apptronik builds bodies and collects deployment data, Google supplies the intelligence layer.

Brain vs Body: Which Strategy Reaches Your Living Room First?

Google's approach reaches real environments sooner because it rides on hardware that already exists. But it inherits every limitation of its partners' bodies — and none of those partners are selling a home robot today either.

OpenAI's vertical bet could eventually produce a tightly integrated product, the way owning both hardware and software worked for the iPhone. But humanoid hardware has killed optimistic timelines before — including OpenAI's own 2021 exit.

What our database says about who is actually closest to home

Strip away the lab announcements and look at purchasable or pre-orderable home humanoids in the ui44 database:

• 1X NEO ($20,000 Early Access, $499/month subscription tier planned) remains the only home-focused humanoid with public pricing and a 2026 delivery window. Soft 30 kg body, 167 cm tall, designed for safe coexistence.
• Apptronik Apollo and Apollo 2 are enterprise platforms with no public pricing — Mercedes-Benz factory pilots, not living rooms.
• Figure 03 is pursuing the same vertical integration OpenAI now wants, but its autonomy claims deserve the skepticism we documented earlier — and Figure at least has physical robots iterating in warehouses, which is more than a hiring page.

Neither OpenAI nor Google changes this ranking today. Google's licensing model at least flows through Apollo 2, a deployed platform. OpenAI's body does not exist yet.

The deployment evidence gap is worth spelling out. NEO's home claim rests on three concrete things: public pricing ($20,000 Early Access plus a planned $499/month tier), a stated 2026 delivery window, and a body explicitly designed for coexistence — 30 kg soft construction at 167 cm, light enough that a bump is a scare rather than an injury. No other home-focused humanoid in our database currently combines all three. Apollo 2, by contrast, has stronger enterprise deployment evidence — an operating Robot Park in Austin, paying factory pilots such as Mercedes-Benz, and a teleoperation data pipeline that feeds the DeepMind collaboration directly. That makes Apollo 2 the better technology bet today, but it is not a home product: no public pricing, no consumer safety case, no living-room navigation story.

So what should a buyer actually do with these two announcements? First, treat them as timelines, not products. If you are considering a NEO preorder, neither the OpenAI confirmation nor Gemini Robotics 2 changes the calculus — NEO's delivery risk is about 1X's manufacturing and support, not about whose lab strategy wins. Second, if you are waiting for a cheaper or smarter option, Google's licensing path is the one to watch: an Apollo-class body running Gemini On-Device 2 that adapts to a new home layout with only a few hours of data is a plausible 2027–2028 enterprise-to-home story, whereas an OpenAI-built home body has no public milestone to track at all. Third, keep an eye on the data loop each strategy needs. OpenAI must build a fleet before it can learn from one; Google learns from every partner deployment but must keep those partners aligned. The first strategy to show weekly deployment hours in real homes — not demo videos — is the one to take seriously.

The honest takeaway

There is one more reason the licensing model matters for homes specifically: cost structure. A licensable brain amortizes its R&D across every partner chassis, so the intelligence layer gets cheaper as more bodies adopt it — the same economics that put Android in budget phones. A vertically integrated body, by contrast, must recover its full hardware R&D from its own unit sales, which pushes early pricing up (NEO's $20,000 Early Access tag is a preview of that math). If OpenAI follows the same path, expect its first robot to be expensive and enterprise-leaning regardless of long-term home ambitions.

Google is winning the short game: Gemini Robotics 2 is shippable software running on real robots now, and the on-device fast-adaptation story directly addresses the hardest home problem — every living room is different.

OpenAI is playing a longer game: owning the body means owning the data loop, and physical interaction data is the scarcest input in robotics. If actuators and manufacturing go well, a vertically integrated OpenAI home robot could leapfrog licensed-brain competitors — in the late 2020s, not next year.

For anyone deciding whether to buy a home humanoid today: neither announcement should change your plans. Watch 1X NEO deliveries and Apollo 2's DeepMind collaboration for real deployment evidence. Lab strategies are interesting; shipping robots are what matter.

Sources: Forbes (Sept 3, 2026) on Altman's Sources podcast confirmation and OpenAI actuator hiring; Humanoid Guide summary of the confirmation; Google DeepMind's Gemini Robotics 2 announcement (July 30, 2026). Robot specs and pricing from the ui44 database, last verified as noted on each robot page.