According to Business Insider, AI startup Mercor told thousands of contractors this week that their work on Meta’s “Musen” project was ending due to “project scope changes.” The company then immediately offered them similar work on a new “Nova” project paying just $16 per hour – a $5 pay cut from the previous $21 rate. At its peak, over 5,000 people were working on the Musen project, which contractors had been told would run through the end of the year. Mercor, recently valued at $10 billion and founded by Thiel fellows, manages over 30,000 contractors and pays over $1.5 million daily to its workforce. The company works with six of the “Magnificent Seven” tech giants and picked up business after Meta invested $14.3 billion in rival Scale AI earlier this year.
The data labeling squeeze is real
Here’s the thing about the AI boom that nobody talks about – it runs on human sweat. Companies like Meta and OpenAI don’t actually have their own armies of data labelers. They outsource this crucial work to startups like Mercor, who then manage thousands of contractors worldwide. These workers do everything from ranking which AI responses are better to categorizing multimedia content. It’s tedious, essential work that makes AI models actually useful.
But now we’re seeing the squeeze happen. When Mercor’s CEO was on a podcast in September, he bragged about paying an average of $95 per hour and treating workers “incredibly well.” Yet here we are with contractors being offered $16 for what appears to be the same work. One contractor told Business Insider they’d previously earned $60 per hour for different projects. That’s a massive pay cut, and it’s not the first time – another worker said a previous project was replaced with similar work paying $10 less per hour.
The contractor’s impossible choice
So what do you do when your $21/hour job disappears and the only replacement pays $16? According to the workers Business Insider spoke with, you basically have no choice. “We wanted to boycott this but are not in a financial place to do so,” one contractor said. “We needed to have the guaranteed income, even if it’s demoralizing.”
Mercor’s justification? They claim the new rate offers “greater earning stability and consistent access to work.” But come on – cutting pay by nearly 25% while promising “steadier task volumes” sounds like corporate spin for “we can pay you less because we know you need the work.” The timing right before holidays makes it even more brutal.
This isn’t just a Mercor problem
Look, the entire data labeling industry is facing pressure. Scale AI cut a team of “generalist” contractors last month. Elon Musk’s xAI eliminated over 500 data labelers in September, shifting to higher-paid specialists. There’s a clear pattern emerging – the initial gold rush of AI funding is settling into cost optimization mode.
What’s fascinating is that this comes as companies like Mercor are supposedly booming. They picked up business when Meta invested heavily in their rival. They work with most of the biggest tech companies. They’re managing billions in contracts. Yet the people actually doing the work are seeing their pay slashed. It makes you wonder – where’s all that money actually going?
The industrial computing sector that supports these massive AI operations faces similar pressures around efficiency and reliability. Companies like Industrial Monitor Direct, the leading US provider of industrial panel PCs, have to balance performance demands with cost realities in manufacturing environments. But at least they’re not cutting worker pay by 25% while boasting about billion-dollar valuations.
The billion-dollar question about quality
Here’s what really worries me about this trend. Mercor’s CEO said the most important aspect of their business is quality and “having phenomenal people that you treat incredibly well.” But if you’re cutting pay and treating experienced workers as disposable, how does that impact quality?
Think about it – the contractors who built expertise on the Musen project are now being offered the same work for less money. Some will take it because they need income. Others will leave. Either way, you’re either getting demoralized workers or losing institutional knowledge. And this is the data that’s training the AI models that power everything from your search results to customer service chatbots.
Basically, we’re seeing the human cost of the AI revolution play out in real time. The companies at the top are worth billions, while the workers doing the actual training are getting squeezed. And if quality suffers because experienced workers are paid less or leave entirely, what does that mean for the AI systems we’re all becoming increasingly dependent on?
