Who Owns the Machine? : The Algorithm #3
Seven signals from Africa, India, and Latin America on who builds AI and how it impacts the communities building around it.
The Algorithm
AI is being built, regulated, and resisted across the Global South at a pace that rarely makes it into mainstream tech coverage. Each week, The Algorithm tracks the most important developments: new models, policy moves, funding gaps, and the harder questions about who AI is actually for.
This week’s take
The story of AI in the Global South is still, at its core, a supply chain story. That is what this fortnight keeps coming back to, across every region.
In Nairobi, over a thousand workers who spent years labelling data, the invisible labour that makes AI systems function, were let go when a contract between two American companies ended. In Andhra Pradesh, farmers are watching their land get acquired under “public purpose” laws so that Google and Microsoft can build the data centres that will power AI for the world. In India’s tribal regions, languages are being digitised by government AI projects without any framework for the communities those languages belong to.
The raw material is being extracted. Most of the value is being created somewhere else.
And then, in the same fortnight: a Bengaluru startup raises $350 million and becomes a unicorn. Morocco signs a billion-dollar infrastructure deal. South Africa publishes a thoughtful, autonomous AI policy.
The Global South is not passive in this story, it is building, regulating, and pushing back. But the tension between who is building and who is bearing the cost runs through all of it.
1. India’s sovereign AI bet just hit unicorn territory
Inc42 / Bloomberg, April 18, 2026 | Sarvam AI is closing a $300–350 million funding round at a valuation of approximately $1.5 billion, with participation from Bessemer Venture Partners, Nvidia, and Amazon. The Bengaluru-based startup builds full-stack AI infrastructure designed specifically for India’s 22 official languages, positioning itself as the country’s most credible domestic answer to GPT-4 and its successors.
This round makes Sarvam the first clear unicorn to emerge from India’s sovereign AI push and a reference point for other Global South nations asking whether you can build serious AI infrastructure outside the US-China duopoly.
2. Over 1100 Kenyan AI data workers lose their jobs as Meta ends a contract
TechCabal, April 16, 2026 | Sama, the US-based “impact sourcing” company, laid off 1108 employees at its Nairobi hub after Meta terminated its content moderation and AI data labeling contract. These workers had been annotating training data for Meta’s AI systems, including its smart glasses. The mass layoff puts a sharp point on a question this newsletter has been asking for a while: who bears the risk in the global AI supply chain?
The data workers who make AI systems function do so without job security, equity, or visibility the same dynamic visible at the base of every extractive value chain.
3. South Africa chose sector regulators over a single AI authority
TechCabal, April 9, 2026 | South Africa released a new national AI policy framework that deliberately opts for decentralised, sector-specific governance rather than building a new centralised AI regulator. The risk-tiered classification system has surface similarities to the EU AI Act, but the underlying philosophy is different : existing regulators in finance, health, and telecoms retain control, and the policy explicitly prioritises the development of local datasets and African language NLP models.
It is one of the most clearly articulated African departures from the European regulatory template.
4. India is offering Big Tech a 20-year tax holiday to build data centres. Farmers are protesting.
Rest of World, April 13, 2026 | India has rolled out a 20-year tax holiday for foreign cloud providers to accelerate AI data centre construction, attracting billion-dollar commitments from Google and Microsoft in Andhra Pradesh and Telangana. The government is using “public purpose” land acquisition laws to fast-track the projects, which is generating pushback from farmers whose agricultural land sits in the path of the infrastructure build-out.
It is a microcosm of a broader contradiction: India’s digital sovereignty agenda is being physically built on the dispossession of its rural population.
5. Who owns a language? India’s AI expansion has no answer.
Observer Research Foundation, April 20, 2026 | A sharp analysis from ORF identifies a critical gap in India’s AI strategy: flagship government projects like BharatGen and Bhashini are digitising hundreds of languages, including tribal dialects, without any framework for community ownership of that linguistic data. India’s data protection legislation does not address collective rights over language, which means the cultural heritage of indigenous communities can be extracted and commercialised with no mechanism for consent or benefit-sharing.
The piece calls for Language Data Trusts and community-governed oversight. It is the kind of governance question that is invisible until it is too late.
6. Morocco signs a $1.28 billion deal to build an AI factory
TechCabal, April 14, 2026 | As part of its Digital 2030 strategy, Morocco has signed a 12 billion dirham memorandum with UK-based Nexus Core Systems to develop a large-scale AI factory, a high-performance computing centre, AI training hub, and innovation campus all powered by Nvidia infrastructure.
Morocco is positioning itself to become North Africa’s compute hub, in direct competition with Egypt and the Gulf for regional AI infrastructure dominance. The deal is also a reminder that the Global South’s AI ambitions are not uniform: some governments are racing to attract investment and build capacity while others are still debating whether to regulate.
7. Brazil is using AI to clear its notorious tax backlog and it matters beyond the bureaucracy
Valor Internacional, April 14, 2026 | Brazil’s Administrative Council of Tax Appeals has deployed an AI system called Iara to accelerate tax ruling proceedings, with the stated goal of cutting average case duration from nearly four years to two. It is a prosaic story, but it signals something significant: Latin America’s largest economy is quietly embedding AI into its institutional architecture at pace, and the use cases being prioritised are not about productivity theatre, but they are targeting genuinely dysfunctional systems. Brazil’s AI bill, currently moving through Congress, is also being pushed to include voice protection provisions for the country’s significant dubbing industry, under pressure from voice actors organising against AI displacement.
That’s it for this edition of The Algorithm. If you spotted a story we missed — a startup, a policy move, a community using AI in ways nobody’s writing about yet — reply to this email. The best leads come from readers on the ground.
very interesting