Generative artificial intelligence (AI) is sold as the great equaliser. Type a prompt, get an essay, a translation, an image, working code, a plain-language answer to a hard question.
Skills once locked behind years of training are now one click away. It sounds like democratisation.
It is not – not automatically. The AI boom will create staggering wealth. But the bill is already being handed to the people least able to pay it: low-income workers, poorer households, and communities hosting the data centres that make it all run.
Call it what it is – not a digital divide but an AI divide.
Let’s start with the basics. Access to AI is not just an Internet connection. It’s a device. And AI is making devices more expensive.
The chips and memory that power AI data centres are the same components that go into phones and laptops. Demand is up, supply is tight, prices are climbing. A wealthy buyer delays an upgrade. A poor family skips it entirely and gets locked out of banking, healthcare, government services and jobs that now assume a smartphone in every pocket.
Then come the jobs. AI is marketed as a tool that “augments” creative and strategic work.
Fine, if you already have creative or strategic work. Most underprivileged workers don’t. They are in retail, admin, customer service, transport and manufacturing – the routine, repetitive roles that automation eats first.
The World Bank’s 2026 assessment is blunt – AI could widen the gap between rich and poor countries because computing power, data and skills are already unevenly distributed. It’s not just that jobs disappear. It’s that the entry-level ladder into better-paid work gets kicked away.
This lands hardest where it can least be absorbed. Asia and Africa are adding huge numbers of working-age people every year. If AI shrinks entry-level work faster than economies create new opportunities, the fallout won’t stay economic. It will be social, and it will be significant.
Then there’s the price of admission. Through 2026, “free” AI increasingly means capped, throttled and stripped of the features that actually matter. Usage-based pricing is squeezing even deep-pocketed players: Microsoft reportedly pulled developer access to Claude Code months after rolling it out, and Uber blew through its entire 2026 AI coding budget by April.
If billion-dollar companies are rationing AI use, what chance does a student in a low-income country have of getting real value from it, rather than a watered-down demo? The gap between “AI as a headline” and “AI as a usable tool” isn’t closing. It’s widening.
The environmental bill is just as lopsided. A United Nations (UN) University assessment from June 2026 projects that AI data centres could consume 945 terawatt-hours of electricity a year by 2030, and 9.3 trillion litres of water. That’s enough to cover the basic domestic water needs of 1.3 billion people in Sub-Saharan Africa.
The communities hosting this infrastructure rarely capture the profits it generates. They just inherit the strain on the grid and the water table.
And then there’s fraud. Generative AI has made scams cheap and scalable: convincing phishing, cloned voices, fabricated video, all achievable without real technical skill. A financial loss that’s an inconvenience for a wealthy household can be catastrophic for a poor one. The cost of attacking someone is falling faster than most people’s ability to defend against it. The poor face a double hit: less access to AI’s upside, more exposure to its downside.
So ask the real questions. Can low-income families afford the devices AI now assumes they own? Can poor students access capable AI without paying subscriptions they can’t afford?
Do displaced workers have any realistic path into new jobs? Are the communities hosting data centres protected, or just squeezed? Is anyone measuring the environmental costs honestly, let alone distributing them fairly?
And the biggest question of all: who actually captures the productivity gains?
If the answer is tech companies, investors and already-skilled workers, while everyone else eats higher device prices, job disruption, strained infrastructure, environmental damage and rising fraud, then this isn’t democratisation. It’s extraction with better marketing.
None of this means slowing AI down. It means building it deliberately. Governments need to invest in affordable connectivity, reskilling and social protection. Companies need to reckon with the social and environmental costs of the infrastructure they depend on. Policymakers need to make sure gains are shared, not hoarded by those who already hold the capital, the compute and the skills.
AI can genuinely democratise access to knowledge and capability. It can also, just as easily, deepen every inequality we already have. The difference isn’t the technology. It’s whether we choose to act.
We should not stop the AI revolution. We should make it inclusive.
Professor Dr Selvakumar Manickam is the director of the Cybersecurity Research Centre, Universiti Sains Malaysia.
The views expressed here are the personal opinion of the writer and do not represent those of Twentytwo13.
