Hidden workforce behind AI boom: how Madagascar's low-wage workers train the world's algor

Nearly two billion people worldwide use artificial intelligence in 2025. None of it works without human hands.

Behind every algorithm lies the labor of thousands of annotators, low-wage workers who train AI systems by labeling images, classifying videos, and tagging text. To cut costs, major AI companies outsource this work to the cheapest markets available. Africa has become a prime target. Madagascar, in particular, has emerged as a critical hub for this invisible workforce, and the consequences for ordinary Malagasy citizens are both significant and largely hidden from view.

The island nation has transformed into the fourth-largest outsourcing destination for French companies, with more than thirty annotation firms now operating in the capital, Antananarivo. The sector generates over 73 million dollars annually and represents more than three percent of Madagascar’s GDP. Yet behind those figures lies a troubling reality: the workers performing these essential tasks face precarious conditions, minimal pay, and limited legal protection.

Madagascar’s appeal to international AI companies rests on several converging factors. Sixty percent of the population is under age 25, and the labor market is saturated with unemployed and underemployed workers. The workforce is francophone, a significant advantage for French companies. Most critically, wages remain extremely low. Annotators earn approximately 60 euros per month for forty hours of weekly work, roughly six cents per click. Many work in large open offices housing up to 1,000 people, subject to strict rules and constant managerial surveillance. The profit margins are stark: while a worker earns perhaps 100 euros monthly, companies resell that annotated data for 1,000 to 2,000 euros.

The workforce itself reveals a deeper problem. Contrary to assumptions that annotation attracts only unskilled workers, research shows that 75 percent of annotators have attended university. Nurses, accountants, and other trained professionals perform these repetitive tasks because Madagascar’s formal job market offers no alternative. A young person with professional qualifications faces no better option than clicking through images for subsistence wages. The mismatch between education and employment reflects systemic economic dysfunction, not individual failure.

Informal work compounds the precarity. Some annotators work through online platforms from home, using personal computers purchased from previous informal income, stacking multiple jobs to stay afloat. Even when companies formally employ workers, many remain undeclared and therefore ineligible for public assistance or social services. Amazon, concerned about its public image, has excluded Madagascar from its official outsourcing list, yet Malagasy workers continue performing tasks for the company through indirect channels, operating entirely outside any legal framework.

Meanwhile, the Malagasy government is navigating a difficult balance between attracting investment and protecting its citizens. Minister of Digital Development Stephanie Delmotte has pointed to the country’s young population and low labor costs, with a legal minimum wage of approximately 55 euros monthly, as draws for foreign firms. She has also signaled that formalizing the sector requires developing legal frameworks and incentives rather than punitive measures, an approach that acknowledges the government’s duty to workers without simply driving companies away.

Some firms have begun charting a different course. ArkeUp, a French digital solutions company and one of the earliest investors in Madagascar’s annotation sector, pays annotators roughly 120 euros monthly, double the average, and provides health insurance and free transportation. The company prioritizes hiring for more qualified positions, directly addressing the brain drain that has long plagued the country. Telesourcia, one of the few Malagasy-owned annotation companies, formally declares its 300 employees and pays around 100 euros monthly. Having evolved from a call center operation in 2015, Telesourcia pivoted to annotation in 2017 and now ranks among the sector’s leaders despite stiff competition from foreign firms.

These examples demonstrate that a more responsible model is achievable. The question is whether it scales.

Madagascar’s annotation sector illustrates both the promise and the peril of AI’s global expansion. For a young population facing a saturated formal labor market, annotation work offers real employment and income. Yet it also deepens precarity, suppresses wages, and leaves most workers outside the legal protections that citizens in wealthier countries take for granted. The government’s challenge, and the country’s genuine opportunity, lies in transforming annotation from a survival strategy into a sustainable engine of economic development and skills advancement. Whether that shift happens depends on the choices made by international companies, the Malagasy state, and emerging local firms in the years immediately ahead.