Every few weeks a story lands that makes you re-check the numbers. This week it's OpenAI's robotics hiring spree: 27 open robotics-related job listings, up from just 11 four months ago, with base salaries reaching $500,000 a year. That's roughly ₹4.2 crore — for an individual engineer. But here's the thing I want you to hold onto while everyone else gawks at the salary figure: the hiring isn't really the story. What those job listings reveal, once you map them onto OpenAI's two robotics facilities, is something much bigger — an industrial pipeline for manufacturing the one ingredient that has held general-purpose robotics back since the beginning.
01.Labeled data — the boring bottleneck
Modern robots learn from what researchers call state-action datasets: recordings of a human hand doing a task, paired with exactly what the robot's joints should do at every moment. Want a robot that can make toast, fold clothes, or plug in a cable? You need thousands of hours of humans demonstrating it, captured in machine-learnable form. Nobody has enough of it. That's why a robot can land a rocket booster but still fumbles a doorknob.
OpenAI's answer is refreshingly unglamorous: build a data factory. In San Francisco, it runs a teleoperation facility that launched in February, pairing GELLO controllers with Franka robotic arms. It operates 24 hours a day across three shifts, with contract workers directing the arms through progressively complex household tasks — yes, including toasters — to generate exactly those state-action datasets. Since launch, the workforce has grown from roughly 100 contract workers to more than 400. That's not a research lab. That's a production line.
02.A warehouse with a data center's appetite
The second facility is where it gets wild. In March 2026, OpenAI signed a lease for 202,000 square feet (about 18,766 square metres) of industrial space at the Portside Commerce Center in Richmond, California — a 15.75-acre waterfront site. County property records confirm the lease was signed on March 9 by OpenAI OpCo LLC. Here's the detail engineers will appreciate: the building's electrical capacity exceeds 14,000 amperes, versus the 200–4,000 amps of a typical industrial warehouse. Your house probably runs on 200. That power density is comparable to a small data center — because that's essentially what it is, except instead of processing tokens, it feeds arrays of robotic test cells, actuation rigs, and the GPU clusters that turn raw teleoperation footage into trained robot policies.
03.The race OpenAI is entering late
Context matters here, and OpenAI is not early. Google DeepMind released Gemini Robotics 2 on July 30 — a family of three vision-language-action models that can control a humanoid's whole body (legs, torso, arms and hands) under a single trained policy, demonstrated on Apptronik's Apollo 2 robot. Its published benchmarks show a 68.4% success rate on tabletop pickup tasks and 45.7% on floor-level pickup — numbers that show both how far the field has come and how far it still has to go before you'd trust it in a kitchen. Meanwhile, Tesla is converting its Fremont factory from Model S/X production to Optimus humanoid manufacturing, targeting a million units a year at scale.
All three companies are chasing the same approach: large models that take visual input and natural-language instructions and output motor commands. We've covered the humanoid money pouring into this space before — from D-Robotics' $400M humanoid chip venture to Figure's Helix 2.5 doing household chores. OpenAI's twist is that it's building the training data infrastructure from scratch — which gives it direct control of the data moat, but puts it years behind rivals in deployed hardware experience.
04.What Sam Altman wants out of this
In a May 31 post announcing the robotics push, Sam Altman described the near-term target as robots that support skilled workers building infrastructure, with data centers named as one of the first application domains. His bigger vision: a world where everyone has their own personal robot doing anything they need. The post picked up 158,000 views, and honestly, the data-center angle is the part I'd watch — it pairs neatly with the AI infrastructure buildout everyone's racing to fund.
It hasn't been a smooth ride, though. In March, Caitlin Kalinowski — the former Meta AR hardware lead OpenAI hired in late 2024 to run its robotics and consumer hardware division — resigned over OpenAI's decision to deploy its models on the Pentagon's classified cloud network (a deal reported at around $200 million), writing that surveillance without judicial oversight and lethal autonomy without human authorization "deserved more deliberation than they got." The ambition and the controversy are arriving together, as they tend to.
05.Why this matters beyond Silicon Valley
Here's my honest take: the salary number will age fast, but the structure of this story won't. When the most valuable AI company on earth decides that the scarce resource is human-demonstrated training data, it tells you where the next wave of work is. Teleoperation and data-collection facilities like the San Francisco lab are labour-intensive by design — and that's a kind of work that could plausibly happen anywhere with reliable power and internet, including India. The bigger prize is the layer above it: people who know how to direct AI systems — break a physical task down, orchestrate the data pipeline, manage compute costs — are becoming the new bottleneck. That's the same skill set our GPT-6 Astra — Zero to Hero Guide builds from scratch, especially its chapters on AI agents and cost optimization — because whether you're training robots or chatbots, the discipline of not burning money on compute is identical. And if this story has you rethinking where your own career fits, our AI Job Hunt Kit — 60 copy-paste prompts for resumes, LinkedIn and salary negotiation, built for the Indian job market — is a practical place to start.
Key Takeaways
- OpenAI posted 27 robotics jobs this week (up from 11), with base salaries up to $500,000 — about ₹4.2 crore a year
- Its San Francisco teleoperation lab runs GELLO controllers with Franka arms 24/7 across three shifts, staffed by 400+ contract workers generating state-action training data
- A 202,000 sq ft Richmond warehouse with 14,000+ amps of electrical capacity functions like a small data center for robot training
- Google DeepMind's Gemini Robotics 2 (July 30) showed 68.4% tabletop pickup success on Apptronik's Apollo 2; Tesla targets 1M Optimus units a year
- OpenAI's bet: controlling the training data pipeline matters more than being first to ship hardware
Frequently Asked Questions
How much are OpenAI's robotics jobs paying?
OpenAI posted 27 robotics-related job listings in the week covered, up from 11 four months earlier, with base salaries reaching $500,000 a year — roughly ₹4.2 crore for an individual engineer.
What does OpenAI's teleoperation lab do?
OpenAI's San Francisco teleoperation facility, launched in February, pairs GELLO controllers with Franka robotic arms and runs 24 hours a day across three shifts. Contract workers guide the arms through everyday tasks like making toast to produce state-action training datasets. The workforce has grown from roughly 100 to more than 400 people.
Why did OpenAI lease a 202,000 sq ft warehouse?
In March 2026 OpenAI signed a lease for 202,000 square feet at the Portside Commerce Center in Richmond, California. The building's electrical capacity exceeds 14,000 amperes — comparable to a small data center — to power robotic test cells, actuation rigs and the GPU clusters that turn teleoperation footage into trained robot policies.
Who is OpenAI competing against in robotics?
Google DeepMind released Gemini Robotics 2 on July 30, 2026 — vision-language-action models demonstrated on Apptronik's Apollo 2 robot, with a 68.4% success rate on tabletop pickup tasks and 45.7% on floor-level pickup. Tesla is converting its Fremont factory to manufacture Optimus humanoids, targeting a million units a year at scale.
What This Means for You
For Indian engineers and students, don't just see the ₹4.2 crore salary — see what it's paying for. It's not roboticists in the classic sense; it's people who understand AI models deeply enough to build data pipelines around them. That's a stack you can start learning today, entirely from a laptop: prompting, agent orchestration, cost control, and safety practices. The GPT-6 era skills in our GPT-6 Astra — Zero to Hero Guide translate directly to the robotics wave, because the models orchestrating those robot arms are the same family of models you'd be directing in any other domain.
One more thing — if you are still getting up to speed on GPT-6 Astra itself, our complete GPT-6 Astra walkthrough explains the model, the pricing and the privacy basics in one sitting.