Some AI companies are trying to literally clean up how

Older readers may remember that during the dot-com bubble, companies offered all sorts of things like Internet connections, CDs, storage, financial services, transportation, housing, and e-commerce. The logic has always been the same: lose money today to build a dominant position tomorrow in a market whose size, in many cases, was more a guess than a reality.

Many companies disappeared when reality took over PowerPoint presentations. Others survived and became giants. But they all share one characteristic: the conviction that growth justifies any level of subsidy.

Alarmingly parallel numbers have been emerging for two decades in the AI industry.

take it shiftA startup that has begun Offer free apartment cleaning in New York. At first sight, it seems crazy. A company sends someone to clean your house, doesn’t charge you anything, and walks away. In a city like Manhattan, where a typical cleaning service can easily cost between $150 and $300, the offer seems to defy any basic economic logic.

Obviously there is a catch. Or, more precisely, a business model: Shift cleaners clean while wearing a camera that records all activity from a first-person perspective. Every movement, every object handled, every decision to clean a messy kitchen or a dirty bathroom is recorded and then anonymized. The customer gets a clean house, and transfers something even more valuable: training data for AI robotic systems.

The math adds up: A cleaner costs about $18 an hour plus coordination, equipment, insurance, decontamination and management costs; Let’s say each session may cost the company roughly $45. From a traditional perspective, it looks like a cost. From a data economics perspective, this is a bargain.

The reason is simple: training data for robotics has become one of the scarcest resources in the entire AI industry. While large language models have been able to glean large amounts of text from the Internet for years, robots need something more complex: observing how humans interact with the physical world.

Cleaning a countertop, picking up clothes off the floor, unloading a dishwasher or moving objects in tight spaces are seemingly trivial tasks for one person. For a robot, they represent something of the toughest. They involve spatial perception, planning, adapting to unexpected environments, and physical manipulation of objects that are never the same.

This is why robotic training can cost anywhere from $100 to $500 for an hour of annotated video. This is why specialized datasets for robotic manipulation typically cost between $50,000 and $200,000. And that’s why a market worth about $753 million in 2024 is projected to reach $6.75 billion in 2031, an annual growth rate of close to 37%.

Seen this way, offering a cleaning session doesn’t seem so crazy.

If a session generates several hours of useful data for training a robotic system, the value obtained may outweigh the cost of the service provided. Free cleaning products are not. You are the product. Or, more accurately, your household’s activities have been converted into data

Are there any differences from the business model that defines the Internet era? Actually, very little: Google has used search to build an advertising empire. Facebook has given up on social media to capture attention and personal information. Uber has been subsidizing it for years to win the market. Amazon has sacrificed margins for decades to achieve scale.

Now we’re seeing companies offering physical services because the data needed to train next-generation AI systems has enormous potential value. The point is, much of that value remains speculative, as was the case during the dot-com bubble.

No one knows for sure which companies will dominate the robotics economy of the future. No one knows which models will ultimately prevail. No one knows whether future revenues will truly justify the current investment. The only thing that seems clear is that there is a mad race to acquire assets considered strategic, and those assets are data.

In this context, free apartment cleaning in Manhattan ceases to be a curiosity and becomes a symptom. A symptom of an industry that is beginning to behave like the Internet companies of the late 1990s: capitalizing on something they hope to monetize in the future.

History teaches us that some of these players will build giant companies. It also teaches us that many more will end up as mere footnotes in subsequent bubble bursts.

Because when a company knocks on your door to clean your house for free, it’s probably not in the cleaning business: it’s probably in the business of trying to buy the future. And, just as it was twenty-five years ago, no one still knows how much it’s worth.

–

This post was Previously published on Enrique Dans’ blog.

***