Beijing’s tech firms exploit crippling unemployment crisis to power artificial intelligence ambitions through opaque networks paying workers less than $6 a day
It is three o’clock in the morning in Nairobi. Ken sits hunched over his laptop in the darkness, eyes flickering between his phone and computer screen.
He has been working for nine hours straight, watching the same 10-second video clips on repeat, trying to determine whether beach waves are crashing in slow motion or a woman is stretching into yoga poses at normal speed.
His WhatsApp buzzes.
A teammate has already labelled 2,200 video clips that day. She is exhausted. Ken still has hundreds more to go before he can sleep. Tomorrow, he will wake up and do it all again. For this gruelling work, which can stretch to 12 hours a day, seven days a week, he earns 700 Kenyan shillings. Roughly $5.42.
Ken is one of thousands of young Kenyans quietly powering China’s artificial intelligence revolution from makeshift digital sweatshops operating entirely through WhatsApp groups and shadowy middlemen.
While American tech giants like Meta and OpenAI have faced mounting scrutiny over their exploitation of African data workers, Chinese AI companies have slipped into Kenya through the back door, building an empire on even more precarious terms with virtually no accountability.
The arrangement represents what labour rights activists are calling a new form of digital colonialism.
Unlike their Western counterparts, Chinese firms operate through deliberately opaque supply chains that make it nearly impossible to trace which companies are benefiting from the labour or hold anyone accountable when workers are exploited.
An Invisible Workforce
None of the 10 Kenyan data annotators interviewed for this story knew the names of the Chinese companies behind the projects they worked on.
They knew only their immediate supervisors and the anonymous portals where they submitted their work, platforms like Vranno.ai that open to nothing more than a login page with no publicly available information about ownership or operations.
Workers are recruited through a simple Google Form, managed entirely through WhatsApp groups of up to 30 members, and paid through the local mobile money service M-Pesa.
There are no formal contracts, no human resources departments, no office buildings. Just a phone number, a group chat, and the constant pressure to label faster, work longer, maintain perfect accuracy or face being cut from the project entirely.
“We just get on the platform where some Chinese managers organize the work,” said David, a university student who has been doing this work for three months and asked to use a pseudonym to protect his income. “We have no idea what they are doing with this annotation work.”
This opacity is by design, according to experts tracking the global AI supply chain.
Chinese AI firms have become among the world’s largest buyers of human-labelled data, but unlike increasingly scrutinized American operations, they work through layers of subcontractors that obscure the trail.
“What distinguishes their expansion is not just scale, but opacity,” said Payal Arora, professor of inclusive AI culture at Utrecht University in the Netherlands. The lack of transparency means far less is known about labour conditions, wage structures or worker protections than with Western firms.
Chinese AI companies contacted for this story did not respond to requests for comment about their operations in Kenya.
Digital Factory Floors
The WhatsApp groups function like digital assembly lines. Every morning, administrators post production targets and daily rankings comparing each worker’s output and accuracy.
Multiple times per week, supervisors hold video calls to review performance reports, flag errors, and push teams to work faster. When one worker falls behind, others are ordered to pick up the slack.
The pressure is relentless.
Teams typically go through a trial period where they must collectively label 20,000 video clips per day with at least 90 percent accuracy.
A single person falling below standards can get the entire team fired.
After passing this simulation phase, individuals are expected to annotate up to 26,000 videos daily, work that can take 12 hours for beginners.
Experienced annotators like Ken have learned to split their screens, using both phone and computer simultaneously, recognizing patterns to speed through the mind-numbing work. “You get into the zone and zone out. You become a zombie,” he said. “If I stop, I lag. If I think, I fail.”
Payment requires maintaining at least 85 percent accuracy.
There is no room for error, no tolerance for the fatigue that inevitably sets in after hours of staring at screens.
Fertile Ground for Exploitation
Chinese firms have found extraordinarily fertile ground in Kenya.
Youth unemployment in the country has reached a staggering 67 percent, according to the Federation of Kenya Employers.
More than one million young people enter the labour market annually, many with university degrees but no prospects. In this environment, even exploitative work becomes attractive.
“Kenya hits all the top spots for global outsourcing,” said Shikoh Gitau, founder of Nairobi-based IT provider Qhala.
“Language, literacy, power stability, and a tech-savvy population familiar with Western culture. Our time zone works magic. We can work with the West Coast of the US and the east coast of Asia without much adjustment.”
But Kenya’s advantages for tech companies translate directly into vulnerabilities for workers.
The country’s current labour laws, designed for traditional employment, offer no protections for digital gig workers.
Meanwhile, the government has been slow to formulate regulations, missing its own July deadline to finalize a framework that would establish who bears responsibility for these workers, the platforms or the companies contracting them.
“A lot of work is going on around firmly identifying who should be held accountable as the employers of these workers,” said Florence Kimata, a member of Kenya’s National Innovation Technical Committee. That work remains unfinished.
The China Model
The exploitation of Kenyan workers mirrors practices Chinese AI companies have perfected at home. A 2023 investigation found that Chinese firms employed vast armies of low-wage data annotators recruited from vocational schools or funnelled through labelling centres in impoverished provinces like Gansu, Guizhou, and Henan to keep costs down and scale quickly.
Now they have exported that model abroad, but with even less oversight.
The business model relies on what Joan Kinyua, president of the Data Labelers Association, a Nairobi-based workers’ union, calls “distance and deniability.”
“It is capitalism and the height of digital colonialism,” Kinyua said. “Oftentimes, supervisors do not even mention who you are working for, but you would be able to tell from the faces in the content.”
One Kenyan supervisor who ran a team of 30 workers was blunt about the economic calculation.
“The bigger the project, the more people we hire and the lower the rates we offer,” the supervisor said. “We cannot have people full-time with employment benefits, so we set that expectation.”
Projects typically last only two weeks, ensuring workers never gain enough stability to organize or demand better conditions.
When projects end, workers are cut loose without notice, left to scramble for the next opportunity.
Old Economics, New Technology
The AI industry may present itself as futuristic, but it runs on profoundly old-world economics.
Behind every sleek chatbot and autonomous vehicle are armies of poorly paid workers doing psychologically draining tasks for a few dollars a day.
“Models look automated, but behind the scenes, they are propped up by armies of low-paid workers,” Arora said.
“Companies rely on cheap annotation not because it is optional, but because the current AI business model depends on absorbing massive training costs while still competing on speed. Cheap labour is the silent subsidy keeping the AI boom afloat.”
The global data annotation market, currently valued at $2.56 billion, is expected to grow at 18 percent annually to exceed $13 billion by 2034.
That growth will be built on the backs of workers like Ken and David, labouring in the shadows while tech giants reap billions.
“They are all here simply because of cheap labour,” Kinyua said.
“Companies know Kenyans will give them quality work done at very minimal or zero cost. At times, these workers do not get paid because they do not have a direct link with the organization they have been working for.”
Without contracts, workers have no recourse when payment is delayed or never arrives. It becomes their word against a faceless platform that can simply disappear, as American company Scale AI’s Remotasks platform did in Kenya in March 2024, abandoning thousands of workers with just hours’ notice.
A Race to the Bottom
The Chinese entry into Kenya’s data labelling market threatens to accelerate a race to the bottom that has already devastated workers across the Global South.
Venezuela, devastated by economic collapse, saw data labelling wages plummet to as low as 90 cents per hour as desperate workers flooded platforms. When conditions improved slightly, companies simply moved operations to countries with cheaper labour.
The pattern is repeating in Kenya and across East Africa, Southeast Asia, and the Middle East.
Companies shift from market to market, always seeking the most vulnerable populations willing to work for the least money.
The lack of local infrastructure requirements makes it trivially easy. A WhatsApp group can be created in minutes. When workers start demanding better conditions, the operation moves to the next country overnight.
Kenya’s position in this exploitative system is particularly precarious.
The country has invested heavily in positioning itself as Africa’s tech hub, branding itself “Silicon Savannah.”
But without strong labour protections or transparency requirements, that brand serves primarily to attract companies looking for educated workers they can pay poverty wages.
Activists warn that without urgent action, Kenya and other African countries risk becoming permanent sites of extraction in the global AI economy, providing the raw labour that powers technologies they will never afford to develop themselves, let alone benefit from.
“AI may feel futuristic, but it is built on profoundly old-world economics,” Arora said. “Without fair labour practices, the future of AI will be fast, but fundamentally unjust.”
At three o’clock in the morning, Ken is still labelling videos. The sun will rise in a few hours. He will sleep briefly, then wake up and begin again. The Chinese AI companies whose models he is training will never know his name. He will never know theirs. That is precisely how the system is designed to work.
Additional reporting by Rest of World
