We’re being sold a peculiar future. The story goes like this: Build larger frontier AI models, put them everywhere, make them do everything. Cover the planet (and then space) with data centers. In this dystopian future, we are burning barrels of fuel to micro-dose on dopamine from synthetic intimacy.
Writer Chimamanda Ngozi Adichie warned us about “the danger of a single story”—how flattening a people, a place, or an idea into one narrative robs it of complexity, and ultimately of truth. The dominant AI story is about creating superhuman intelligence (the one that knows everything). The alternative AI story that barely grabs headlines is about distributing human agency (help everyone with what they need).
RENTED SOVEREIGNTY
Consider what full-stack AI sovereignty looks like through this dominant story lens: Saudi Arabia is spending $100 billion, mostly with U.S. tech companies, to build 11 data centers, 2,200 megawatts, hundreds of thousands of chips, and its own Arabic-native LLM. The ambition is real, but the framing gets complicated when the infrastructure is built on Nvidia chips and Google Cloud. This version of sovereignty is also unattainable for most nations.
Enter Amini, an AI infrastructure stack designed specifically to address barriers to AI participation: fragmented data, limited connectivity, and scarce compute. Amini converts paper records into machine-readable intelligence and deploys portable, decentralized microdata centers designed for local ownership and interoperability. Barbados got its data centers shipped in containers and operationalized the system in six months, initially launching with a 0.1 megawatt system. Citizens now access government services through a multilingual WhatsApp assistant. The containers belong to the Barbadian government. The data, too.
KETCHUP AND HOT SAUCE
There’s a concept in Hindi, jugaad, that denotesfrugal innovation, and its parallel in Swahili, jua kali. The insight is that constraint isn’t a limitation; it molds the design. The current LLM moment is ketchup: poured on everything, applied indiscriminately. Every query we run draws on the entire corpus of digitized (scraped) human knowledge, with all the biases this entails. Frugal AI is hot sauce: applied strategically, in the right dose, to the right problem. Models small enough to run on microcontrollers. Systems that work offline and on edge devices. Intelligence adapted to solving specific localized problems.
Rology deploys AI cancer diagnostics in communities that have no radiologists, across Kenya, Egypt, and Saudi Arabia. The model is small enough to run locally and fine-tune affordably. Fetosense, out of Mumbai, is a portable device that allows midwives to detect fetal distress. It includes a clinically validated algorithm that helps triage cases to specialists. True value is in delivering useful intelligence where it is scarce.
OWN THE UPSIDE
Communities across the world are already inside the AI economy, just not as beneficiaries. Data workers in Kenya, the Philippines, and India do the annotation and labeling that make frontier models function, with none of the upside. They are ghost workers in the latest iteration of a supply chain that extracts value and routes it elsewhere.
Contrast that with what Awana Digital (formerly Mapeo) has built with the Siekopai people of the Amazon: a locally-owned database co-created with the community to document and reclaim 42,000 hectares of ancestral land. The Siekopai own their data, and now their land—AI facilitating restitution rather than extraction.
THE STORIES WE TELL
“Humans in the loop of AI” is a common trope, but a misdirect, like a hamster spinning on its stationary wheel. Motion without direction, activity without agency. Taiwan Cyber Ambassador, Audrey Tang, suggests a reframe: AI in the loop of communities, and of humanity.
The dominant AI narrative is seductive: bigger, faster, inevitable. It is the plausible world, barreling towards singularity: the extrapolation of what already exists, who already has power, what’s already funded. The possible world of plurality is within our reach—in containers in Barbados, ancestral land records in the Amazon, and local compute in a clinic in Cairo. AI in the loop of humanity.
The danger of a single story is that it monopolizes our imagination, crowding out alternatives before they can prove their value. The future should be measured not by how much intelligence we can concentrate in one system, but by how much agency we can distribute across humanity.
That is the story still waiting to be told.
Hala Hanna is the executive director of MIT Solve.
No comments