Every argument about the data centre — the jobs, the water, the gas, the site at Debe or Point Lisas — collapses into one question that decides all the others: do we own it, or are we hosting someone else's? Answer that, and the whole cost-benefit calculation flips. So let us actually do the sum, honestly, both ways.
Our position has been consistent across this whole series. A sovereign, state-owned data centre — powered by our own gas, owned by the people — is one of the best industrial decisions this country could make. Build that one tomorrow. But the deal actually on the table is not that. It is three Memoranda of Understanding with US organisations — Ernst & Young LLP and Hummingbird AI Holdings LLC — to build up to 800 MW of foreign-owned compute, brokered, in the government's own words, with the "facilitating" hand of Washington. Those are two completely different projects wearing the same press release. This article costs out both.
If the data centre is ours, the benefits are ours. If it belongs to a US corporation, we supply the costs and they collect the benefits. That is the entire analysis in two sentences.
Trinidad and Tobago Socialist Party
The Benefits They Are Selling
The government's pitch is three headline numbers: US$5 billion in potential investment, over 5,000 jobs, and the vague promise of putting Trinidad "on the map" as a Caribbean AI hub. It sounds enormous. So let us take each benefit and weigh it against what it actually costs us to obtain it — because a benefit you pay full price for, and then hand the profit to someone else, is not a benefit. It is a purchase.
Benefit #1: "5,000 Jobs" — The Construction Mirage
This is the number that sells the deal, and it is the most misleading. There are two kinds of data-centre jobs, and they are worlds apart. Construction jobs are real but temporary — a burst of hiring while the shell goes up, then the crew is laid off. Permanent operational jobs are what you actually live on afterwards, and hyperscale data centres are engineered to need almost none of them.
The industry's own figures are brutal. Operations typically run at 0.15 to 0.35 permanent staff per megawatt — some highly automated hyperscale sites run on as few as 20 to 30 people per 100 MW. Do the arithmetic on our 800 MW: that is somewhere around 150 to 280 permanent jobs once the diggers leave. Not 5,000. A few hundred — security guards, a handful of technicians, some management. In Northern Virginia, the data-centre capital of the world, the facilities generate roughly one permanent job for every US$13 million invested, against about US$137,000 to create a job in most other sectors — nearly one hundred times less employment per dollar. Analysts have a phrase for what these deals really trade in: not jobs, but joules.
So the honest translation of "5,000 jobs" is: a two-to-three-year construction boom, followed by a bust. The scaffolding comes down, the workers go home, and the country is left with a hard-guarded building that employs a few hundred people and hums quietly while shipping its value abroad. Ask any community that was promised a factory would "transform the area" and got a warehouse with a car park. Then what? is the question no minister answers, because the answer is: not much.
The Cost Side: What Trinidad Actually Pays
Now the column nobody in government reads aloud. To land a foreign data centre, host governments hand over a stack of subsidies — and every one of them is a real cost to the Treasury and to citizens:
• Subsidised gas and electricity. The single biggest running cost of an AI data centre is power. Our advantage is cheap gas — but if we sell that power to a foreign owner at a discounted, subsidised rate, we are simply exporting our gas at a loss and calling it investment. Every megawatt burned for their servers is a megawatt of our finite resource we did not sell at market, or use for ourselves.
• Tax holidays. These projects live in "special economic zones" precisely so they can be granted years — often decades — of tax exemptions. Reduced or zero corporation tax means the "US$5 billion investment" generates a fraction of the public revenue the headline implies.
• Free or cheap state land. Serviced, zoned land handed over at concessionary rates.
• Water and grid infrastructure — built by us. The government has already admitted it must build man-made ponds and desalination plants and harden the grid to feed the site. That is public money spent so a private foreign company can operate.
• The environmental bill. Heat, water strain, and emissions — paid by the community, collected by the corporation.
Add it up and the shape becomes clear: we supply the gas, the land, the water, the grid, the tax break and the environmental risk. They supply the servers — and keep the profit. That is not an investment in Trinidad. It is an extraction from Trinidad, structured as a favour.
Where Is the Grift?
Ask the Trini question directly: where in this deal will politicians and their friends try to eat? A project this big, this opaque, moving this fast, is a buffet. Watch these plates:
• The middleman structure. Ernst & Young is a consulting firm, not a data-centre operator — the MOU says it "intends to partner with third parties." Who are the third parties? Hummingbird AI Holdings is an LLC that can keep its real backers behind a single name. Undisclosed beneficial owners are where kickbacks hide. Internationally, most jurisdictions that hand out data-centre subsidies do not even disclose which companies receive them — secrecy is the industry norm, and secrecy is the grifter's oxygen.
• The construction contracts. The one genuinely large, genuinely local flow of money here is the build — earthworks, concrete, steel, electrical. That is exactly where the contract mafia feeds: inflated tenders, no-bid awards, and companies that materialise the week the MOU is signed.
• The land. Who owns the acreage being rezoned into a "special economic zone," and who bought in early?
• The utility side deals. Ponds, desalination plants, grid upgrades — each is its own contract, its own opportunity.
None of this can be policed while the actual contract stays secret. That is why our first demand has never changed: publish the deal in full. Ownership, the tax holidays, the gas price, the land terms, the beneficial owners — all of it. A deal that cannot survive daylight is telling you what it is.
The Environmental Cost — Paid in Full, Here
The environment is where "hosting someone else's data centre" gets its worst bargain, because the damage cannot be exported the way the profit can. The heat stays here. As we detailed on the Debe siting, data centres run their surroundings measurably hotter — creating localised heat islands — while their thirst for cooling water lands in a country that already rations supply. And there is a deeper irony: we would be burning our own finite natural gas, and emitting the carbon, so that a US company can train and sell AI to the world. We take the emissions and the depleted reservoir; they take the model and the revenue. If the compute were ours, at least the country that pays the environmental price would own the thing the price bought. Hosting it for someone else, we pay the bill and receive a receipt.
The Scorecard: Sovereign vs Foreign-Owned
Here is the whole analysis on one page. Same gas, same land, same water, same building — two completely different outcomes depending on one word: ownership.
| Question | Our Sovereign Data Centre | Foreign US-Owned Data Centre |
|---|---|---|
| Who owns the compute? | The people of T&T | A US corporation |
| Where do the profits go? | Circulate at home, in local hands | Repatriated to Wall Street in US$ |
| Permanent jobs | A few hundred — plus the engineers, researchers and AI industry built on top of it | A few hundred. Full stop. |
| Our cheap gas buys… | National value, kept here | A discount for a foreign owner |
| Tax holidays & free land | N/A — the state is the owner | Given away; little public revenue |
| The environmental bill | Paid for something we own | Paid for something we don't |
| Data sovereignty | Our data under our law | Under foreign law & subpoenas |
| Can we build our own AI/LLMs? | Yes — the whole point | No — we rent access like everyone else |
| Is it worth it? | YES — build it now | NO — not on these terms |
The Prize We Give Up: Our Own AI
Here is the benefit that never appears in the government's slide deck, because a foreign-owned deal quietly deletes it. If the compute is ours, the data centre is not the product — it is the factory. With sovereign processing power we can train our own large language models: a Caribbean AI that actually understands our dialect, our history, our law, our context — not a Silicon Valley model that thinks "lime" is only a fruit. We can point that compute at the things that matter: hospital diagnostics, agricultural planning, flood modelling, education, research, government services. We can sell sovereign, low-latency cloud to the rest of CARICOM in our own currency. That is how a country actually makes money from compute — by owning the machine and selling what it produces, not by renting out the shed it sits in.
A foreign-owned data centre gives us none of that. We would still be renting AI access from a US corporation — except now the servers sit on our land, burn our gas, and heat our air while we do it. We would have paid the full cost of hosting the future and kept none of it. This is the difference between building a factory and letting someone build a factory on your lawn that you are then charged to use.
The Verdict: Is It Worth It?
So — is it worth it? The only honest answer is: it depends entirely on whose name is on the deed.
If it is ours, the sum is overwhelmingly positive. A short-term construction boom, yes — but followed by a permanent national asset: sovereign compute, our own AI industry, data security, value that compounds at home for decades, and a reason for our best engineers to stop leaving. Every cost we pay buys something we own. Build it. Now.
If it belongs to a US corporation, the sum is a trap dressed as a gift. We pay in gas, land, water, tax revenue, environment and grid; we receive a construction boom that ends, a few hundred permanent jobs, and a building that ships its profit and its power abroad. When the ribbon-cutting photos fade and the workers are laid off, the honest question returns and stays unanswered: then what? On those terms, it is not worth it — not because we fear the technology, but because we refuse to pay full price to be a tenant on our own soil. That is not development. It is digital colonialism with a local flag on the gate.
We are not anti-data-centre. We are anti-giveaway. Publish the contract, guarantee public ownership and control, site it correctly at Point Lisas rather than Debe, and turn this from an extraction into an asset. Own the compute, own the future. Rent it, and we own nothing at all.