Demand for low-cost human labelers is rapidly declining as AI models mature and require more sophisticated data.
Top artificial intelligence firms are shifting away from low-cost data labelers in Africa and Asia. Instead, they are hiring highly paid domain experts to build more powerful and intelligent models. Companies like Scale AI, Turing, and Toloka are now seeking professionals in fields such as biology, finance, physics, and software engineering to craft the next generation of AI training datasets. This pivot is driven by the emergence of advanced “reasoning” models, including OpenAI’s o3 and Google’s Gemini 2.5, which require increasingly complex and high-quality data.
Shifting Landscape for AI Data Labeling
For years, the AI industry relied heavily on gig economy workers. In countries like Kenya and the Philippines, data labelers earned less than $2 per hour for repetitive tasks. These included drawing bounding boxes around images, filtering graphic content, and refining phrasing. These workers formed the invisible backbone of AI development, often under pressure to complete hundreds of microtasks daily.
“The AI industry was for a long time heavily focused on the models and compute, and data has always been an overseen part of AI,” stated Megorskaya, CEO. “Finally, [the industry] is accepting the importance of the data for training.”
With basic annotation tasks becoming increasingly automated, the need for low-cost human labelers is sharply decreasing. AI companies are now investing in specialists. These experts curate nuanced datasets tailored for domain-specific reasoning. AI firms now require professionals who can demonstrate chain-of-thought reasoning, solve real-world problems step-by-step, and simulate complex scientific theories. For instance, a physicist might design a theoretical experiment, a software engineer would code a simulator to test it, and a data scientist would analyze the output.
“The result of this is the model’s not just going to be better than a physicist. It’s going to be better than a superposition of somebody who’s at the top in physics, computer science and data science,” explained Jonathan Siddharth, co-founder and CEO of Turing AI.
Experienced software engineers are also tasked with creating domain-relevant problems, solving them by writing and debugging code, and assessing outcomes for security risks. This strategic shift has spurred significant investor interest. In June, Meta invested $15 billion into Scale AI, doubling its valuation to $29 billion. Earlier, in March, Turing AI secured $111 million at a $2.2 billion valuation. In May, Jeff Bezos’ investment firm, Bezos Expeditions, led a $72 million funding round for Toloka. These investments highlight a growing conviction that superior data, not just advanced models, will be the key differentiator in the competitive AI landscape.
While approximately 10-15% of AI budgets are allocated to data, the immense scale of AI investments means these figures still represent “enormous” sums, Siddharth noted. Despite the declining demand for simpler tasks, some opportunities remain for gig workers. Joan Kinyua, president of the Data Labellers Association in Kenya, reported that local workers are now focusing on tasks requiring localized language knowledge. Additionally, some human labelers continue to perform final quality control checks, evaluating and validating AI-generated content.
