Meta’s AI Restructuring Is Erasing the Nairobi Jobs That Helped Build It

On April 16, 2026, Samasource Impact Sourcing Inc issued formal redundancy notices to 1,108 workers at its Nairobi delivery center after Meta Platforms terminated a major data-annotation and content-processing contract. The engagement had been central to Sama’s Kenyan operation for years. Thirty-four days later, on May 20, Meta began notifying roughly 8,000 of its own global employees that they were being cut, part of a restructuring the company says is necessary to fund an AI infrastructure program running at up to $145 billion in capital expenditure this year.

Both decisions belong to the same logic. Workers in Nairobi spent years labeling images, moderating harmful video, and processing data streams that trained Meta’s AI systems. Those systems are now the primary reason Meta says it needs fewer humans. That circuit, from outsourced labor in Kenya to in-house automation in California, has run to its conclusion at both ends simultaneously.

Nairobi’s Workforce and the Contract That Powered It

Kenya’s emergence as a digital-outsourcing hub was not accidental. The government invested in fiber-optic connectivity, English-language education, and policies designed to attract global technology companies seeking lower labor costs. By 2025, Kenya’s business-process outsourcing sector (BPO, meaning contracted digital services delivered remotely for international clients) was valued at roughly $270 million annually, employing around 36,000 workers in roles spanning customer support, financial processing, and AI data services, according to industry figures compiled by Kenyan investment media.

Sama occupied a prominent position in that ecosystem. The San Francisco-headquartered firm promoted itself as an “impact sourcing” operation, deliberately placing workers from underserved communities into digital roles connected to global tech clients. Meta contracted Sama in 2019 to moderate Facebook content across sub-Saharan Africa, a task requiring local language knowledge and cultural judgment that automated systems of the era could not handle. Over years of mounting legal disputes and labor complaints, Sama exited direct content moderation and pivoted its Kenyan workforce into AI data annotation, the systematic labeling of images, video, and audio that large-scale machine-learning systems require at every stage of development. That AI-labeling workstream is what Meta has just ended.

The formal redundancy process is being conducted under Section 40 of Kenya’s Employment Act, which requires written notification to employees, the labor office, and relevant trade unions, with a minimum severance of 15 days per year of service. Sama said it had engaged Meta in an attempt to preserve the contract but acknowledged those discussions were unsuccessful. A Swedish media investigation by Svenska Dagbladet and Göteborgs-Posten, conducted alongside Kenyan journalists and published in the weeks before the notices arrived, found that footage captured by users of Meta’s AI glasses was being reviewed and labeled by contracted workers in Nairobi to train Meta’s underlying models. The investigation raised questions about user consent and the accountability distance between the platform and the people doing its most sensitive processing work that went publicly unanswered as the workforce handling that review was being dissolved.

For many of the workers now without contracts, the relationship had paid meaningfully above Kenya’s formal minimum wage but placed every strategic decision, including the decision to end it, exclusively with the foreign client.

What Sama Workers Were Building for Meta

Sama’s Nairobi center ran two related streams of work. Content moderation required workers to review graphic violence, hate speech, and extremist material flagged on Facebook across African markets. AI data annotation involved labeling images, video clips, and audio samples so Meta’s machine-learning systems could be trained to perform those judgments automatically. Both types of work required sustained human attention. Both fed the same AI pipeline that Meta is now funding at a scale that renders the human version economically redundant.

The financial terms of that arrangement have been disputed since the relationship began. Several independent assessments establish the structure of the compensation.

• 2.5× the formal minimum wage: average compensation found by a J-PAL evaluation of Sama’s Nairobi delivery center, with benefits including healthcare, pension, and meal subsidies
• $414/month: what some former moderators told The Associated Press they were paid while reviewing traumatic content for eight-hour daily shifts
• $2.20/hour: the hourly rate described by former moderator Daniel Motaung, who reviewed beheadings and child abuse material under a Sama contract for Meta

The gap between those wages and the scale of the infrastructure they supported is not incidental. A 2026 peer-reviewed study presented at the ACM Conference on Human Factors in Computing Systems, based on interviews with Kenyan data workers, described their position as a regime of entrapment in Kenya’s AI supply chain, held in place by short-term contracts, non-disclosure agreements, and a structural barrier to moving into higher-value roles. The workers who labeled Meta’s training data were, in aggregate, subsidizing an AI buildout in which they held no equity and over which they had no say.

Sama, not Meta, was the legal employer of record throughout. That structure meant Meta could close the contract without the Nairobi job losses appearing anywhere on its own headcount figures. The 1,108 workers simply ceased to be a line item.

A Legal Fight That Has Outlasted Two Contracts

Kenya’s courts have been working through the consequences of Meta’s outsourcing model since 2022. In March 2023, 43 former content moderators, backed by Foxglove, a U.K.-based technology rights nonprofit, filed suit alleging Meta had directed its replacement contractor, Luxembourg firm Majorel, not to hire former Sama employees, treating the contract cancellation as retaliation for worker organizing rather than genuine redundancy. A separate case involving 185 moderators from across African markets sought $1.6 billion in compensation for psychological harm and the absence of adequate safeguards against the cumulative toll of years spent reviewing violent imagery. Settlement talks opened by court order collapsed in October 2023, with Foxglove describing Meta’s conduct in those negotiations as insincere. Meta argued throughout that Kenyan courts had no jurisdiction over a company not incorporated in Kenya and that the moderators were employees of Sama, not of Meta itself.

As is standard in our industry, client programmes evolve, and we work closely with our partners to manage these transitions responsibly.

Annepeace Alwala, Sama’s country lead and vice-president for global delivery, offered that statement when the April 2026 redundancy notices were filed. It mirrors language the company used in earlier contract transitions. The Kenyan judiciary has been less willing to accept the underlying framing. In 2024, Kenya’s Court of Appeal ruled that Meta could be sued within Kenya, rejecting years of jurisdictional arguments. The 185-moderator case is now proceeding on that basis, with liability claims against a company whose newest round of contract terminations has just added several hundred more potential plaintiffs to the broader conversation about working conditions in its outsourced ecosystem.

Meta’s Financial Equation

The Numbers Behind the Restructuring

Meta’s decision to cut roughly 8,000 direct employees globally while accelerating AI infrastructure spending reflects a specific corporate arithmetic. Analysts at Evercore estimate the May round will generate approximately $3 billion in annualized savings, against a capital expenditure guidance range for 2026 that runs between $115 billion and $145 billion, nearly double the $72.2 billion the company spent in 2025. The savings, in other words, cover roughly two cents of every dollar the company plans to spend on AI infrastructure this year.

The May round is the largest companywide reduction since Mark Zuckerberg’s “Year of Efficiency” campaign in 2022 and 2023, which removed roughly 21,000 positions. Meta’s full-year revenue crossed $200 billion for the first time in 2025. These cuts are not a response to financial pressure; they are a reallocation of cost from people to machines, and the savings they generate are a rounding error against the infrastructure bill they are meant to partially offset.

The AI Infrastructure Replacing Those Roles

In June 2025, Meta hired Alexandr Wang, the former chief executive of Scale AI, as its first chief AI officer through a transaction that included a $14.3 billion investment in Scale AI. Wang now leads Meta Superintelligence Labs, the division that debuted its first major model in early May. Alongside that organizational shift, the company is building Prometheus, a one-gigawatt AI supercluster in Ohio, and Hyperion, a 2,250-acre facility in Louisiana rated for five gigawatts of AI computing capacity.

Scale AI, before Wang’s departure, was itself a significant buyer of outsourced data-labeling work from contractors in Kenya and comparable lower-cost markets. Meta’s move toward absorbing that capability in-house describes precisely what happened to Sama at the bottom of the same supply chain, only one level up. The company that organized the labelers has been acquired by the platform. The workers who did the actual labeling were not part of that transaction, and there is no equivalent mechanism for them to follow the work upward.

The Automation Floor Under Kenya’s Outsourcing Sector

The Sama layoffs are the most visible single event in a structural problem that has been accumulating for years. Kenya’s government has publicly set a target of 1 million BPO jobs within five years and secured a $10 million investment pledge from ADEC Innovations at the Kenya International Investment Conference held weeks before the redundancy notices arrived. The aggregate numbers for the sector look optimistic. The contract-level reality is more fragile.

• Concentration risk: A small number of U.S. tech companies account for a disproportionate share of Kenya’s BPO volume; the Sama-Meta contract showed how a single termination can remove over a thousand positions overnight with no domestic absorber capable of replacing the demand.
• Automation exposure: Research by Caribou Digital and Genesis Analytics, conducted with support from the Mastercard Foundation, found 40% of tasks in Africa’s tech outsourcing sector could be automated by 2030, with customer-experience roles (44% of current sector employment) carrying the highest near-term risk.
• Gender disparity: The same research found tasks performed by women are on average 10% more vulnerable to automation than tasks performed by men in the same sector, concentrating displacement in the workforce segment with the fewest alternative pathways into higher-value digital roles.
• Value-chain position: Kenya ranks 11th globally as an outsourcing destination by the 2026 Ataraxis Global Outsourcing Talent Index, a genuine competitive strength; it does not address the fundamental issue that every contract renewal or cancellation is decided by a handful of foreign clients whose own technology strategies, not Kenya’s workforce policy, determine how many contracts exist.

A 2026 CHI Conference paper drawn from interviews with Kenyan data workers described the sector’s core tension in terms the workers themselves had identified: the AI models that required human training data to become functional are, once functional, being deployed to remove the humans who generated that data from the next version of the pipeline. The workers being let go helped build the reason they are being let go.

Two outcomes will determine whether the Sama layoffs become a turning point or a template. Kenya’s appellate courts now have jurisdiction over Meta, and the 185-moderator case is proceeding. If those plaintiffs win meaningful compensation, every global platform that has outsourced sensitive AI labor to African contractors will need to recalculate how it structures liability and what counts as an employment relationship. If the case resolves without consequence, the model that built Kenya’s digital-labor economy, and that is now eroding it, carries on under exactly the same terms as before.