Europe has AI rules. Africa has AI rules. We have a recommendation.

By Amplifi AI · Caribbean AI Brief · Issue Nº 08 · Weeks of 20 July to 2 August 2026 You know the road. Somebody came and sprayed white marks on it. Little arrows, numbers, a code nobody could read. Everybody in the district said the works were finally coming. That was two years ago. The paint has faded. The road is the same. Nothing was wrong with the survey. The survey was excellent. Somewhere between the survey and the asphalt, the thing died. I want you to hold that picture, because two weeks ago the Caribbean produced something genuinely important, and whether it matters at all depends entirely on whether we have learned anything about the gap between a plan and a road. Here is where we currently stand. Europe has an AI Act. The African Union has a continental AI strategy. CARICOM has, as of two weeks ago, a very good recommendation that we ought to get something. But the recommendation is genuinely excellent, and the reason it is excellent is worth your next five minutes.

The region finally has an AI plan

I told you in the last brief to follow the Caribbean AI Forum. Here is what came out of it. On 23 and 24 July at UWI St. Augustine, the Caribbean Telecommunications Union, The UWI and the Artificial Intelligence Innovation Centre held the first Caribbean Artificial Intelligence Forum. Ministers, regulators, diplomats, academics and private sector people in one room. The forum is not the story. The document is. The Caribbean AI Task Force released its final report, **Toward Harmonized AI Policies and Recommendations for the Caribbean**. Thirty six members. A year of consultation. Chaired by Dr. Craig Ramlal of the AI Innovation Centre. Seven strategic recommendations covering data and digital infrastructure, governance and ethics, economic transformation, talent and skills, AI in the public sector, climate resilience, and a coordinated Caribbean voice in global AI governance. That is the first time this region has put a single agreed position on AI on paper. Not a workshop summary. Not an endorsement of somebody else's roadmap. Ours.

Two phrases worth carrying into every meeting you attend this year

Most of the report will not stay with you. Two phrases from the launch should. **The coordination gap.** Twenty odd territories writing twenty odd sets of rules, or in most cases none at all. If you operate in Barbados, Jamaica and Guyana today, you are guessing at three different answers to the same question about what you may legally do with customer data. **The implementation deficit.** We produce excellent strategy. We are considerably weaker at the part where strategy becomes a thing a citizen can touch. That is the white paint on the road. Dr. Ramlal put the underlying maths plainly. Individually our markets are too small. Individually our skilled people are spread too thin. Individually the financial, computational and institutional burden is too heavy. The minimum workable scale for fixing this is regional, not national. Which brings us back to the line I opened with. **CARICOM is currently the only comparable regional bloc in the world without an agreed collective AI governance instrument.** Sit with where that was said. Not by a critic outside the region. Not by a foreign consultant. It was said out loud at the launch of our own report, by the people who wrote it. That is what makes this document different from every regional strategy that came before it. It is not a marketing exercise. It is a group of our own people telling us the truth about ourselves in public. *Our view:* the recommendations are sound and the diagnosis is honest, which is more than most regional documents manage. The test is not the report. The test is whether a funded, time bound implementation mechanism sits behind it before the end of this year, or whether we are back here in 2028 launching the follow up study. Ask your minister which one it is going to be.

The free model arrived. Now read the fine print.

Two weeks ago I told you Moonshot AI would release Kimi K3, one of the most capable models on the planet, free to download on 27 July. I framed it as good news for the Caribbean: every time a world class engine goes free, our ability to build rather than rent improves. It shipped on schedule. The full weights went public on 27 July with the technical report alongside. That part I got right. Here is the part I did not tell you, and you deserve it. Free to download is not free to run. Kimi K3 needs roughly **1.4 terabytes of fast memory** just to hold the model, before it does a single piece of work. Moonshot recommends at least **64 accelerators** for a serious deployment. That is not a laptop. That is not a server in your office. That is a data centre. So who is "open" actually for? Realistically, clouds, inference providers, and governments with the budget to keep 1.4 terabytes warm and waiting. Which is more or less the same short list that already runs everything else at this scale. **This is the whole Caribbean story in miniature.** The software is free. The infrastructure is not. Access was never the bottleneck. It never is. What it changes for you this month is almost nothing. You will still reach these models the way you already do. What it changes over five years is who is capable of running the hardware, and whether any of them are here. Which brings us neatly to Trinidad.

The money is still moving faster than the rules

The report was still being written when Trinidad and Tobago signed its data centre agreements with US firms, a package the government values at over US$5 billion, the first of their kind with any Caribbean nation. Put the two facts side by side and the picture sharpens. The capital is arriving before the governance is settled. That is not automatically a disaster. It does mean the terms of this region's participation in the AI economy are being written right now, in rooms, by whoever turned up prepared. And a signed MOU is not an operational campus. The distance between the two is measured in megawatts, permits, water, skilled staff and legal clarity. Somebody finances that distance. We should all want to know who, and on what terms.

What this means for your business

**1. The window for input is open and it will not stay open.** The Blue Ribbon Commission, the Task Force recommendations, the national policies being drafted right now. These processes decide procurement rules, data rules and training funding for the next decade. Caribbean operators and entrepreneurs are barely in these rooms. If you have ever complained that policy is written by people who have never run a business, this is the year to stop complaining and send the email. **2. Stop waiting for infrastructure to save you.** The lesson of Kimi K3 is that the tool being free changes very little about your Tuesday. What changes your Tuesday is whether your team knows how to use what is already sitting on their laptops. That gap is not closed by a data centre. It is closed by training, and only you can decide to start. **3. Watch for the funding language to shift.** Skills, governance and local ownership are now stated regional priorities in a document that will feed the policy agenda. Over the next eighteen months, expect grant calls, procurement requirements and corporate training budgets to start echoing that language. Businesses that already have working AI capability will be the ones positioned to answer.

Coming up

**Ongoing** · The CAITF final report is publicly available from the CTU. If you read one regional document this year, read the recommendations section. It is shorter than you think. **Watch** · Whether COTED attaches an implementation mechanism and a budget to the report, and when. This is the number to hold them to. **Watch** · Whether any Caribbean government moves on compute capacity beyond Trinidad, and whether the Dominican Republic hub stays on track for early 2027. — The fortnight in one sentence: **the region finally wrote the plan, and a world class AI model became free to anyone who can afford a data centre to run it in.** Both stories are the same story. Neither the plan nor the model does anything on its own. Somebody has to pick it up and build. You do not have to solve the region's implementation deficit. But you can refuse to run one inside your own business. Pick the task your team quietly hates. Give it one hour this week. That is the whole job. One hour, repeated. Until next week, Janelle Germain · CEO, Amplifi AI