Sam Altman: OpenAI will 'definitely' build a humanoid ... and he thinks the body is the easy part

OpenAI is making a humanoid robot

Sam Altman said it out loud, and he said it unequivocally. Asked on the Sources podcast whether OpenAI's robotics push meant a humanoid or a data center robot, Altman answered clearly: 

"Both. We will definitely do a humanoid. We will do other form factors as well."

But he thinks the body is relatively easy compared to the AI:

"I think all of that is less important than really figuring out like the the brain that makes the robot work."

That is a thesis. It is also a bet ... one that a very large amount of capital in this sector is currently positioned against. Interestingly, I've asked multiple robotics CEOs what's harder: the body or the brain. Pretty much all of them have said: both.

Just wait until Altman needs a great pair of hands, for instance.

What Altman actually said

The full exchange is worth reading, because the reasoning underneath it is more considered than the usual "the world is built for humans" boilerplate, though he says that too.

"The world is very much designed for people," Altman said. "So if you think about like the ability to open a door and type on a computer and drive a piece of equipment and you know clean a kitchen and whatever else, we've kind of built this world for people and I want to make sure that we keep building this world for people. So matching that form factor seems good."

Then the qualifier: 

"There will of course be data center robots that have like different form factors. I think all of that is less important than really figuring out like the the brain that makes the robot work."

Pressed on whether that means a personal robot for everyone, Altman was more cautious about sequencing than about the destination. 

"I I don't think that's the most important first thing to do," he said — infrastructure and manufacturing come first — "but yes, someday. I think everyone should have a personal robot. Like I would love to have a personal robot that could like do the tasks that I don't want to do. That'd be great."

So: humanoid: yes, form factor: secondary, consumer: eventually, and brain: everything.

"The brain matters more" is a contested claim

Altman is not alone here, and the people who agree with him are serious.

Gill Pratt, CEO of the Toyota Research Institute and the man who ran DARPA's Robotics Challenge, has put it about as directly as it can be put: "What has changed is actually not about humanoids... What's different now isn't the body, but the brain." Pratt's caveat is that we've only solved half the brain — the fast, reflexive System 1 pattern-matching that current policies do well — and that the field risks a trough of disillusionment if nobody cracks System 2 reasoning and world models.

The market has been voting this way with real money. 

Skild AI raised $1.4 billion in January 2026 at a valuation north of $14 billion, led by SoftBank with Nvidia, Bezos Expeditions, Samsung, LG and Salesforce Ventures alongside. Skild builds no robots at all. Its pitch is an "omni-bodied" brain — co-founder Deepak Pathak says it "can control robots it has never trained on, adapting in real time to extreme changes in form or environments."

Physical Intelligence is running the same play, shipping π0.7 in April 2026 as a steerable cross-embodiment foundation model. Google DeepMind shipped Gemini Robotics 2 on July 30, 2026: a whole-body VLA plus an embodied reasoning model, with an on-device variant Google says adapts to a new body within hours. It runs on other people's robots: Apptronik's Apollo 2, Boston Dynamics, Franka, Agile Robots.

And Nvidia, on May 31, 2026 — the same day Altman announced OpenAI Robotics was hiring — published an open reference humanoid design built on a Unitree chassis. Nvidia's message was unsubtle: the body is a commodity we will hand you for free, the brain is where the value is.

Notice who is not building humanoid hardware in that list. Nvidia, DeepMind and Skild have all looked at this market and decided the robot itself is somebody else's problem.

The counter-argument is Elon Musk's factory floor

The brain-first thesis is attractive, but it's not well-supported by fact.

Tesla started producing Optimus at Fremont around August 2026, on converted Model S/X lines, five years after Musk announced the program. Zero customer units have shipped. In January 2026 Musk conceded that no Optimus robots were doing useful work in Tesla's own factories. The January 2025 target of roughly 10,000 units in 2025? Missed by a mile.

Why?

On the most recent earnings call Musk called Optimus "the hardest product to scale manufacturing that we've ever made," said "everything on the robot is new," and that Tesla had to "build up a supply chain in its entirety," describing an S-curve that is "flat and long" at the start.

This is the company that mass-produces some of the most complex consumer hardware on earth saying a humanoid robot body is brutally hard to make.

The supply chain data backs him up.

McKinsey has called the robotics supply chain the most underappreciated constraint on humanoid scale: actuators alone are 40-60% of bill-of-materials cost, harmonic drives and precision roller screws face severe supplier concentration, and China controls roughly 69% of rare-earth mining and about 90% of magnet processing. OpenMind's recent estimate is that building an Optimus Gen 2 without Chinese suppliers would cost roughly three times as much — around $131,000 against a ~$46,000 BOM.

Current humanoid BOMs run $30,000 to $150,000. The mass-market target is under $20,000. McKinsey doesn't expect modularization to inflect until around 2035.

Brett Adcock made the vertical-integration case when Figure walked away from its OpenAI collaboration in February 2025: 

"We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI. We can't outsource AI for the same reason we can't outsource our hardware."

That is a direct rejection of Altman's framing: brain and body have to co-evolve, and you can't decouple them.

OpenAI's job listings might actually disagree with Altman

If the brain is what matters and the body is secondary, the cheapest path is obvious: license someone's chassis, ship your model, skip the five-year manufacturing slog. Nvidia will literally hand you a reference design.

It's not clear OpenAI is doing that, though.

As of September 3, 2026, OpenAI's careers page lists 19 open robotics roles, all in San Francisco — up from 11 at the division's launch on May 31. Four of them are actuator-specific, plus a dedicated technical program manager for actuators: Actuator Design Engineer, Actuator Electromagnetic Design Engineer, Actuator Gear Design Engineer, Electrical Engineer for Actuator Test Infrastructure.

The Actuator Design Engineer listing asks the hire to lead custom electromechanical actuator development end to end — motors, transmissions, sensing, thermal architecture — defined around torque density, efficiency, thermal performance and manufacturability, and to work with manufacturing partners on production-ready designs. 

It pays $225,000 to $423,000 plus equity, four days a week in person.

Nobody designs their own electromagnetics and gear trains because the body is the easy part. Actuators are the single hardest, most expensive, most supply-constrained component in a humanoid. OpenAI is buying the exact expertise that Tesla says is the bottleneck.

There's also a PCB layout engineer, a thermal simulation lead, a firmware engineer, a prototyping lab technician, an inventory manager and a technical commodity manager on that list. 

That's, essentially, the skeleton of a hardware company.

What OpenAI hasn't said

Plenty has been left unsaid, though.

There's no ship date. No launch window. No unit target. No named manufacturing partner. No prototype has been shown publicly, and there's no credible reporting that one exists. Altman gave TIME a concrete timeline in August for the io/Jony Ive tabletop device — early 2027 — and gave the humanoid nothing but "definitely."

OpenAI also hasn't said who runs the thing. 

Caitlin Kalinowski, hired from Meta in November 2024 specifically to restart robotics, resigned on March 7, 2026 over the company's Pentagon deal, three months before the division was formally named. Several outlets have reported that world-simulation lead Aditya Ramesh now runs it, but that appears to be inference rather than confirmation. OpenAI's own pages name no robotics lead.

And there's the history. OpenAI had a robotics team once. Dactyl solved a Rubik's cube one-handed in October 2019. Then co-founder Wojciech Zaremba disbanded the group in 2021, and his stated reason was, precisely, a brain problem: 

"There [are] actually plenty of domains that are very, very rich with data. And ultimately that was holding us back in terms of robotics."

(Which seems to agree with Figure CEO Brett Adcock's position.)

OpenAI quit robotics because it couldn't get data. Nothing about that constraint has been fixed since, and it's why one of the launch reqs was an operations manager for data acquisition, and why Figure is spending a billion dollars doing Index.

The upshot

Altman's brain-over-body claim is potentially defensible. As hard as humanoid robot bodies are, physical AI is really challenging. And if he's right, the hardware-first humanoid companies burning capital on actuator supply chains are building the commodity layer of somebody else's stack.

But OpenAI's hiring says its leadership doesn't believe the strong version of that claim. You don't staff four actuator specializations if the chassis is interchangeable.

The more likely reading is that Altman means something narrower than it sounded: the brain is the harder problem, the one that gates general capability, and the one OpenAI thinks it's uniquely positioned to solve. The body is still hard. It's just hard in a way that money and hiring can eventually grind through ... which is exactly what the job listings are doing.

Meanwhile the shipping happens elsewhere. 

Global humanoid shipments hit roughly 19,100 units in the first half of 2026, up 272% year over year, per Smart Analytics Global. AgiBot took about 8,400 of those and passed Unitree for the top spot. Chinese vendors accounted for more than 97% of global shipments.

None of them are waiting for anyone's brain.

Sources: Sources podcast, OpenAI Robotics careers, OpenAI Actuator Design Engineer listing, TIME, TechCrunch on Kalinowski's resignation, TechCrunch on the Figure split, VentureBeat on the 2021 shutdown, OpenAI's Rubik's Cube post, The Robot Report on Skild, Google DeepMind on Gemini Robotics 2, Nvidia's open humanoid reference design, Physical Intelligence blog, IEEE Spectrum on Gill Pratt, McKinsey on humanoid supply chains, SCMP on H1 2026 shipments