There is a bottleneck at the heart of the robotics revolution, and it has nothing to do with hardware, chips, or funding. It is data. Specifically, the kind of real-world, physical-environment data that teaches robots how to navigate a kitchen, stack a shelf, or fold a towel without breaking something.

Chatbots were trained on the internet. There is no internet for the physical world. And that single fact has created what UC Berkeley professor Ken Goldberg calls the "100,000-year problem": at current data-collection rates, a general-purpose robot built on a ChatGPT-sized training set will not exist for 100,000 years.

Instawork thinks it can close that gap. And the way it plans to do it is by strapping cameras to gig workers.

From Temp Staffing to Robotics Data Factory

Instawork started as a gig-work platform for hourly shifts at hotels, warehouses, and stadiums. It has raised over $150 million from Benchmark, Greylock, and Spark Capital. On Thursday, it announced a full pivot into robotics with the launch of the Instawork Robotics Lab and a robotics certification program that has already reached more than 20,000 workers.

In May, the company plans to release Instacore, a wearable camera system designed in-house. Workers will wear the device during regular shifts in kitchens, warehouses, and hotels, recording how they move, what they touch, and how they adapt when environments change. The data will be sold to robotics companies and researchers training physical AI models.

CEO Sumir Meghani has renamed himself "Chief Robot Officer." That is not a joke. That is on his business card.

The Numbers Are Staggering

Instawork estimates the robotics industry collected roughly 100,000 hours of training data in 2024, 1 million hours in 2025, and is projected to hit 20 million hours in 2026. That sounds like progress until you realize it represents 0.04% of what is needed to close Goldberg's data gap.

Goldman Sachs forecasts the humanoid robot market could reach $38 billion by 2035. The global data-collection and labeling market is projected to hit $17 billion by 2030. The demand for real-world robot training data is enormous and growing. The supply barely exists.

"We Don't Want Humans Simulating Robots"

The key insight behind Instacore is that the data only works if people behave naturally. Aaron Bromberg, Instawork's robotics lead and a former Amazon technologist who worked on the Astro home robot, put it bluntly: "We don't want to record humans simulating robots. We want humans being humans."

The wearable system includes cameras positioned on the body and a small backpack. The company says data collection is opt-in and anonymized, capturing workflows rather than identities. Instawork plans to send thousands of certified workers into real workplaces wearing the device.

The certification program has two tracks: robotics and AI data collection, and certified robot technicians who can handle hardware resets, maintenance, and field support in over 100 markets.

The Paradox at the Center

Here is the thing nobody building these systems wants to say out loud: the robots that will eventually replace human labor in kitchens, warehouses, and factories are being trained by the exact humans who work in those kitchens, warehouses, and factories. The gig economy that was built to fill gaps in human labor is now filling the gap in robot training. The workers are simultaneously the product and the eventual redundancy.

Instawork is positioning this as a new category of work, not a replacement. And in the short term, they are right. Someone has to train the robots. Someone has to maintain them. Someone has to be in the room when they break. But the 100,000-year problem is a countdown, and every hour of data collected makes it shorter.

The question is not whether gig workers will train the robots. They already are. The question is what happens when the robots graduate.

Source: Business Insider