OpenAI Will Build Its Own Humanoid Robot — Demo Due in 2027
By Meta, TechPulse USA
OpenAI has never shipped a physical product. No phone, no speaker, no wearable — the company that put a chatbot in two hundred million pockets did it entirely in software. So when Sam Altman was asked point-blank whether OpenAI would actually build a humanoid robot, the two-word answer landed like a small earthquake in an industry already running on caffeine and capital: "We will."
Here is the twist nobody should miss. Altman does not think the robot is the hard part. He thinks the hard part is everything else about it — the intelligence that tells the body what to do. And he has put a date on it: a demo in 2027, with robots on the streets some years after that. Treat that date as a CEO's target rather than a product plan — but even as a target, it reframes the whole humanoid race. OpenAI is no longer content to rent other people's bodies for its intelligence. It wants its own.
"We will." — How Two Words on a Podcast Broke the News
Altman appeared on the Sources podcast, where he spent most of the conversation making a surprisingly practical case for the human shape. The world, he argued, is "very much designed for people": doors, keyboards, tools, staircases — all of it assumes a body with two arms, two legs, and eyes at the top. A robot meant to move through those spaces without rebuilding them should probably match that form factor. In his words, "matching that form factor seems good."
Then came the direct question — will OpenAI actually build one? Two words: "We will."
The confirmation matters because of who said it and what it reverses. OpenAI spent years as the hardware industry's favourite sugar daddy: backing startups, lending models, letting someone else weld the metal. Building the machine itself is a strategy U-turn.
A Cool Demo in 2027 — and the Street Waits Longer
The timeline arrived in a separate interview. Altman told Fortune: "I think we'll have a cool demo in 2027 ... I don't think there will be robots walking the streets for a few years longer." The quote was relayed by Engadget, which published the story late October 17 (October 18, India time), and independently carried by LiveMint and TokenPost.
Notice what is missing from all of it: what the demo will show, what the machine looks like, and when anyone outside the company gets near one. OpenAI has disclosed none of that. TokenPost's Simon Yoon puts it bluntly — the expected demonstration is not a launch date, and the project remains, in Engadget's words, "some distance from becoming a product." This is a man gesturing at a calendar, not a company announcing a SKU. Hold that distinction; it is the difference between news and noise.
Altman has been seeding this for a while: he told Bloomberg's Emily Chang he does not believe "the world has really had the humanoid robots moment yet" — though it is coming, and will feel "very sci-fi." The 2027 demo is OpenAI's down payment on that moment.
The Brain Is the Bottleneck
Altman says what the robot looks like is "less important than really figuring out ... the brain that makes the robot work." Giving a machine two arms and two legs, he argues, buys you nothing if it does not know what to do with them.
Consider the coffee mug. To pick one up, a useful humanoid must spot the mug, find the handle, estimate the grip, and exert exactly the right force. Every one of those steps is a perception, planning, or control problem; get any of them wrong and you own a coffee spill. Now multiply that by every object in a house, every texture, every wobbling table. That is the workload Altman wants OpenAI doing — and it is, frankly, the workload OpenAI is built for.
OpenAI's Dactyl project taught a robotic hand to manipulate real objects with training done entirely in simulation, then applied the technique to a Rubik's Cube.
It was an early taste of the headache that still plagues every robotics lab on earth: a model can learn what should happen in software, but physics gets the final vote. Simulation-to-reality remains the toll booth every humanoid program must pass through — and Altman is betting OpenAI's models are the fastest way through it.
Read the Job Ads: 27 Roles and Custom Actuators
Do not take Altman's word for the seriousness of this — take the hiring. Engadget counted 27 open robotics roles on OpenAI's careers page, spanning motion control, state estimation, actuator control, whole-robot behaviour, and the machine-learning infrastructure to train physical AI systems. The listings go further than software: they indicate OpenAI is designing custom electromechanical actuators — motors, transmissions, sensing, thermal systems — while other roles cover manufacturing and the supply chain needed to produce hardware at scale.
We checked OpenAI's careers board this morning, and the robotics and actuator-design listings were still live, spanning motion control, state estimation and whole-robot behaviour. That is the detail that should sharpen every reader's focus. A software company does not hire motor people for fun. AI Weekly's analysis of the same postings reached the same conclusion: this reads like a vertically integrated program — in-house design, in-house manufacturing ambition, no outside body to rent.
That is quite a bit of work for the "less important" half of the robot, as Engadget drily notes. Altman may wave away the hardware in interviews — but the payroll tells a different story, and payrolls don't do PR.
First the Factory, Then Your Living Room
One of the postings states the programme's sequencing plainly. The short-term objective: general-purpose robots that help skilled workers build infrastructure. The long-term goal: a personal robot that helps with a much wider variety of tasks. Translation: warehouses and construction sites first, your kitchen much later.
That order is shrewd. A controlled industrial setting tolerates the clumsiness that a living room will not, and it produces the real-world interaction data the "brain" needs to learn. It also dodges the privacy questions — no announced price, no subscription model, no release date, and no robot-specific privacy framework. The moment a camera-and-microphone-equipped machine starts roaming someone's house, those questions get very real, very fast.
The Trail Behind: Dactyl, Figure and 1X
This is not OpenAI's first date with robotics — it is the relationship finally going public. The company disbanded its original robotics team years ago, then spent the 2020s flirting through other people's hardware. It invested in Norway's 1X Technologies through its Startup Fund. In 2024 it signed a partnership pairing OpenAI's models with Figure AI's hardware. It even filed a trademark application with the US Patent Office mentioning "programmable humanoid robots" — a filing is not a product plan, but it is rarely nothing.
Then the partnership strategy started coming apart. Figure walked away in February 2025. CEO Brett Adcock wrote on X that the company had achieved "a major breakthrough on fully end-to-end robot AI, built entirely in-house" — arguing, in effect, that large language models were becoming the smallest piece of the robot puzzle. Meanwhile, The Information reported that OpenAI itself had discussed acquiring 1X outright; the talks did not progress.
The 1X story is not over, either. SoftBank is reportedly in talks to take a majority stake in the company at around $6 billion, per The Information via Reuters relays. OpenAI may have walked away from the acquisition table — but its former dance partner is about to get the richest date in the room.
A Crowded Starting Grid
OpenAI is entering a race where the front-runners already have head starts and scars. Tesla has been working on Optimus for years — Engadget cites it as the incumbent. XPeng's IRON humanoid recently walked off the production line under its own power. Figure is now an independent rival. And per IFR data, 7,000 full-size humanoids had already gone commercial — mostly doing dull, useful things far from any stage light.
Money is flooding in from every direction — DeepSeek's $400M bet on a months-old stealth robot startup is the latest sign that investors believe the bottleneck is about to break.
That last fact is worth sitting with. The humanoid conversation keeps happening as if the category doesn't exist yet. It does — just not in anyone's living room.
Our Take: This Is About Data, Not Debuts
Here is the judgment call this story deserves: the 2027 demo is the least interesting part of this announcement. A controlled demo is choreography — impressive, expensive, and nearly useless as a measure of real capability. The interesting part is the hiring. Twenty-seven robotics roles, custom actuators, manufacturing and supply-chain postings: that is a company building a data pipeline, not a stage act.
Every hour a humanoid spends bumping into a real factory floor generates the physical-AI training data that simulation alone cannot. Startups selling real-world human demonstration data to train robots have already figured out that data, not motors, is the scarce asset. OpenAI — sitting on arguably the best general models on earth and a voracious appetite for training signal — has simply decided to manufacture its own scarcity away. That is a colossally expensive decision, and it tells you what OpenAI believes: that whoever owns the robot owns the data stream, and whoever owns the data stream owns the next decade of physical AI.
The risk is real, and Altman has been honest enough to name the timeline as aspiration. Hardware is an unforgiving business for a company whose culture ships software weekly — actuators don't get patched over the weekend. Whether OpenAI's software edge is big enough to justify learning manufacturing the hard way will decide whether 2027 is a demo or a debutante ball. Either way, the humanoid moment Altman keeps predicting just got its most serious contender.
Frequently Asked Questions
When will OpenAI's humanoid robot be ready?
Altman says OpenAI should have a demo in 2027, but does not expect robots walking the streets for a few years after that. Nothing is announced about the demo's contents, the robot's design, pricing, or public access — treat 2027 as a target, not a launch date.
Why is OpenAI building a humanoid itself instead of partnering?
Altman argues the "brain" — perception, planning, control — matters more than the body, and OpenAI wants that intelligence paired with hardware it controls. It previously backed 1X and partnered with Figure AI; Figure ended that partnership in February 2025, and reported acquisition talks with 1X went nowhere.
What is the main challenge — the body or the brain?
Per Altman, the brain. A humanoid must interpret sensor data, plan tasks, coordinate its body, and recover from mistakes in real time. OpenAI's earlier Dactyl project showed models can learn in simulation — but real-world physics remains the hard part.
Who are OpenAI's competitors in humanoid robots?
Tesla (Optimus), Figure AI, XPeng (IRON), 1X Technologies — reportedly in SoftBank's sights — plus Agility, Unitree and a deep bench of Chinese and American startups. Several thousand full-size humanoids are already working in commercial settings.
What happened to the OpenAI–Figure partnership?
Announced in 2024 to pair OpenAI's models with Figure's hardware, it was ended by Figure's CEO Brett Adcock in February 2025, who said the company's in-house end-to-end robot AI had made the deal unnecessary. OpenAI is now building its own hardware.
Update, Oct. 19: the ‘other form factors’ Altman already promised
On the Sept 2 Sources episode, investor Alex Heath asked whether the hardware push meant a humanoid or a data-centre machine. Altman’s answer: “Both. We will definitely do a humanoid. We will do other form factors as well.” So the robotics programme is bigger than one biped — purpose-built machines, including data-centre hardware, are in scope (tbreak.com, citing the Forbes original; relayed by webpronews.com, aistify.com, caimpare.ai, explainx.ai and robotico.market). The division behind it was announced May 31, 2026, led by Aditya Ramesh, the researcher behind DALL-E (and later Sora). The payroll signal is already visible: the 27 roles Business Insider counted (11 in May, 27 now) carry base pay from $177,000 to $500,000, with inference engineers at $380,000 fixed and senior hardware-engineer total pay near $1.2M per levels.fyi data. Near-term focus is industrial work — including machines to help build and operate data centres — with the personal robot as the longer-term goal. One alternate telling, attributed: AI Weekly characterises the Figure split differently — OpenAI reportedly dissolved the partnership over insufficient physical testing. Our post’s version (Figure’s Adcock walking away on an in-house-AI breakthrough) stands; the competing framing is worth watching as the in-house programme takes shape.
Sources: Engadget ("OpenAI Is Building A Humanoid Robot, But The Body Isn't The Real Challenge," October 17, 2026); LiveMint ("'We will have a cool demo in 2027,'" October 18, 2026); TokenPost ("OpenAI Puts Robot Intelligence at Center of Humanoid Push," October 18, 2026); tbreak ("OpenAI Humanoid Robot: Altman Eyes 2027 Demo"); AI Weekly ("Altman confirms OpenAI will build humanoid, demo in 2027"); The Brief ("OpenAI Confirms It Will Build Its Own Humanoid Robot"); Bloomberg's The Circuit with Emily Chang interview (via transcript); The Information (1X acquisition discussions, SoftBank–1X talks, via multiple relays).
About TechPulse USA: TechPulse USA is an independent technology publication covering the stories shaping the future — artificial intelligence, robotics, and the companies racing to build them. Our reporting is researched from primary sources and reputable outlets, and every fact we print is verified before publication.
Comments
Post a Comment