No, a data centre will not make us AI sovereign (Part 1)

It is a serious question from serious people, and it moves quickly to infrastructure. Where would it go? What would it cost?

Who would build it? Well, I would argue we are getting ahead of ourselves. Digital sovereignty is a country’s ability to govern its data and the systems holding it.

Where may information about citizens sit, who can reach it, whose courts settle a dispute? Questions of location and access. AI sovereignty is a harder proposition.

It is the ability of a nation, government or organisation to maintain complete, independent control over its entire artificial intelligence ecosystem without relying on foreign powers or external vendors. Why harder? Because AI systems no longer just hold information.

They act on it. When software decides who qualifies for a benefit or which child goes into which programme, the man in the street needs to know on what basis, and whether he can argue back. Three things make it worth pursuing: Compliance, meeting our obligations at home and abroad.

Security, since a system hosted overseas sits under a legal order we did not write and cannot appeal. And autonomy — not being thrown off course by geopolitics, vendor lock-in, or a supplier changing its terms on 90 days’ notice. Truth be told, autonomy is the one the private sector feels first and governments notice last.

All of this rests on four pillars, and our position on each differs. Where we actually stand Data. Keeping the information AI systems train on and process in the country, under our laws.

Fully achievable, because we have the instrument. The Data Protection Act 2020 bars the transfer of personal data abroad unless the receiving territory offers adequate protection, with penalties reaching four per cent of worldwide income. What remains open is enforcement.

Can the regulator verify where processing happens, or does it take the sales deck at its word? Operations. Who holds the keys?

Who restores the system after a hurricane or a cyber attack? Access, disaster recovery and availability run locally are a matter of staffing and discipline, not capital expenditure. Low-hanging fruit, if ever I saw it.

Models. Retaining authority over model weights, algorithms and training, rather than renting somebody else’s system. Partly achievable.

Training one from scratch costs billions, but capable models are published openly. Think of it as buying the record rather than streaming the song. Once that file sits on your own servers, nobody abroad can switch it off, and adjusting it with local data and languages is affordable.

Infrastructure. Data centres, networks and the specialised processors AI runs on. This pillar we will not fully hold.

Advanced chips are made in a handful of facilities, overwhelmingly in Taiwan, at a capital intensity no small state will match. And then there is the light bill An AI system is a machine for turning electricity into answers, and power is its largest running cost. Across Caricom states that import fuel, electricity averages 30 to US$0.40 US cents per kilowatt-hour — several times what large industrial users pay in the countries building AI.

Anyone who has opened a Jamaica Public Service bill will understand. We solved connectivity by sharing infrastructure: subsea fibre lands on our shores and capacity spreads across borders at almost no extra cost. Compute cannot be pooled the same way, and the reason is physics, not politics.

Fibre is a thin strand of glass carrying light, light enough to lie on a seabed 6,000 metres down. A power cable is heavy, needs converter stations costing hundreds of millions at either end, and leaks energy over every kilometre. Hence, after decades of feasibility studies, still no regional grid.

Believe it or not, no country satisfies all four. The United States hosts roughly three-quarters of the world’s AI computing power, and nations that built their own facilities to escape American cloud providers simply moved the dependence one link down, from renting American computers to importing American chips. Does that mean we abandon the standard?

Nothing could be further from the truth. We use it as a scorecard, not a pass mark: three pillars within reach, the fourth partly. Which leaves the asymmetry this region has not absorbed.

Data can travel to where the power is. Power cannot travel to where the data is. In Part 2, what that means for how we proceed — and where sovereignty is actually won and lost.

Trevor Forrest is founder and CEO of Lignum Security Limited, a provider of AI-driven cybersecurity solutions and consulting services; member of the National AI Task For ce; and senior advisor for science, technology and special projects in the Office of the Prime Minister.