Robots Still Can't Do Your Dishes — Nvidia Explains the One Big Reason
A robot that cleans your kitchen is close — except for one missing piece.
Nvidia — the company whose chips power most of today's AI — is heading to TechCrunch Disrupt 2026 in San Francisco (October 13–15) to answer a question a lot of people are asking: why aren't robots already doing our chores? Les Karpas, who runs Nvidia's physical AI startup program, will explain in a talk bluntly titled "Robots Are Waiting for Their ChatGPT Moment." Nvidia's CEO Jensen Huang has spent years telling anyone who'll listen that robots are the next big thing. So far, that promise hasn't shown up as a robot in your kitchen.
Karpas's explanation is simple to grasp but brutally hard to fix. When ChatGPT exploded in November 2022, it worked because OpenAI could train it on a massive chunk of the internet — billions of pages of text, all free to use. Robots have no equivalent. There is no internet-wide library of how to fold laundry, unload a dishwasher, or navigate a cluttered garage. Self-driving fleets from Waymo solved a version of this problem by driving millions of real-world miles and collecting every second as data, market by market. Robots haven't logged those miles yet, and there's no shortcut.
A wave of startups is trying to fake that experience through computer simulation — training robots inside digital worlds instead of the real one. It's promising but imperfect: a robot trained in a video game doesn't perfectly know how a slippery plate or a grumpy cat behaves. That gap between the digital and physical world is the core bottleneck Karpas is focused on, and it's why Nvidia created a special "Real World AI" track at the conference, alongside founders from Shield AI, Foxglove, and others.
Why should you care? Because the same data problem is what stands between you and a robot that could cook, clean, or help care for aging parents. It's also where billions in investment are flowing — and where the next big startup fortunes may be made. The catch: this isn't a problem you throw money at and solve overnight. Even optimists say a true breakthrough could take years, and the conference itself is a paid event designed to sell tickets and connections, not give away the answer for free.
- Robots can't learn the way ChatGPT did — there's no giant internet library of real-world movement to train them on.
- Self-driving cars like Waymo got around this by logging millions of real road miles; robots have no equivalent shortcut yet.
- Startups are using computer simulations to fake real-world experience, but simulations don't perfectly match messy reality.
Why It Matters
The robots that could cook, clean, and care for us stay years away until this data problem is solved.