A New Kind of Refinery

The Brief

• On July 10, the Government signed three Memoranda of Understanding with US firms at a US Independence Day event hosted by the US Embassy.
• Ernst & Young LLP: framework for a 300MW data centre using EY's "Energy to Intelligence" platform, with EY to partner with unnamed third parties.
• Hummingbird AI Holdings LLC: framework for a 150MW AI infrastructure facility, with potential expansion to 500MW.
• Pinnacle Steel and Vanadium Corporation: framework for recommissioning the Point Lisas iron and steel plant.
• The Government's combined projection across the three deals is over US$5 billion in investment and more than 5,000 jobs.

The Government has signed three MOUs. The critics have focused on water and energy. I want to spend this post on a different question.

Since the memoranda were signed at the US Embassy on Independence Day, Fortune, AP, the Guardian, the CEO of an AI agency, activist Dr Wayne Kublalsingh, and a 13,000-signature petition have all raised the same set of concerns. Data centres consume vast quantities of water. Trinidad and Tobago has chronic water shortages. Data centres consume vast quantities of electricity. The country's total generation capacity is about 2.4GW, and the three deals combined are proposing to draw up to 800MW of that. These are correct concerns. They are being made repeatedly, from multiple directions, and they deserve serious answers.

What has not been made in any of the coverage I have read is the argument I want to make in this post. It is not about capacity. It is about pattern.

The refinery, in a different shape

The country has been here before.

For most of the second half of the 20th century, Trinidad and Tobago's economic model was extractive. Foreign companies came to the country. They took the raw material, crude oil. They built the physical infrastructure to refine it, right here at Pointe-à-Pierre. They employed local workers, paid royalties to the State, and generated foreign exchange. The country hosted the operation. The value that resulted – petroleum, petrochemicals, refined products for the wider Caribbean market – was owned by the companies operating the facilities. When the industry became unprofitable, in 2018, the refinery closed and the country was left with the debt, the environmental legacy, and the loss of thousands of jobs. The Petrotrin restart, which this blog engaged with back in June, is an attempt to reset that arrangement on more favourable terms. Whether it succeeds is a question the country will answer in the next two or three years.

Hold that history in view, because the shape of what is being proposed for the data centre MOUs is genuinely difficult to see if you look at it only through the environmental lens.

Data centres are not oil refineries. They do not process crude. They do not produce gas at the pump. But they are infrastructure of a specific and specifiable kind. They take a raw material, electricity, and they consume it at industrial scale to power computation. The computation itself is not petroleum. It is artificial intelligence: language models, image generators, agentic systems, cloud services, video processing, financial trading. That computation has enormous value. It is being sold, right now, to enterprises and governments and consumers around the world, generating billions of dollars in revenue per quarter for the operators.

If Trinidad and Tobago hosts a 300MW data centre on the old sugar factory grounds at Usine Ste Madeleine, the country will provide the electricity and the water and the land. What it will not do is train the models, sell the API access, own the customer relationships, or capture the profit. It will collect royalties and taxes, employ a small number of specialist workers, and shoulder the environmental costs of the facility's operation. The country hosts. The value goes elsewhere.

That is the same extractive model that ran through Pointe-à-Pierre for eighty years. It is being proposed again, in a new industry, at what may turn out to be an even larger scale. It is a new kind of refinery.

The pattern of MOUs

Before we get to the substance of the extraction argument, there is a smaller point worth making about how these deals are being announced.

The Petrotrin restart has been under negotiation for well over a year. Minister Moonilal has been to India Energy Week in Goa, to the Guyana Energy Conference in Georgetown, and to Nigerian energy officials. Indian Oil Corporation was expected to visit after Carnival to provide technical advice. Guyana's President Ali connected the delegation with Arab investors. Nigerian entities were consulted. In April, Moonilal told the Express the Government was "nearing a final decision" and that a proposal would be "presented to Cabinet for a final decision in the very near future."

That was three months ago. As of this weekend, no operating partner for Petrotrin has been publicly named. No financing structure has been publicly disclosed. No commercial terms have been announced. The bill establishing the Petrotrin successor structure passed in June, which was a genuine legislative achievement and which I credited at the time. The actual restart has not happened.

I flag this not to criticise Minister Moonilal, whose work on the refinery restart has been serious and sustained, but to name a pattern. This Government, like the several that came before it, is comfortable announcing headline deals at symbolic ceremonies and less comfortable disclosing the operational details that make the difference between an MOU and a functioning facility. The data centre deals are non-binding preliminary frameworks. The Petrotrin restart is a fuller stage of the same journey, and it is not yet at the finish line. The country would be right to be sceptical of the promise timeline on the data centres, given what it has seen with Petrotrin.

The 5,000 jobs claim

The Government's headline claim is that the three MOUs, if realised, will create "more than 5,000 jobs." That figure has been repeated by the Prime Minister, by AP, by international outlets, and by the Government's own press release.

The country deserves a breakdown, and there has not been one.

Here is what we know. Data centres are notoriously low-employment infrastructure. A large data centre, a 200-plus MW facility, typically employs a few dozen to at most a couple of hundred operational staff at steady state. The people who work there are specialist engineers, security personnel, and facilities managers. Construction of the physical facility employs many hundreds of workers over 18 to 36 months, but those jobs are temporary. Once the concrete is poured and the racks are installed, the operational headcount is small. That is by design. Data centres are capital-intensive, not labour-intensive.

Steel plants are different. The Point Lisas iron and steel plant, which Pinnacle Steel and Vanadium Corporation is proposing to recommission, is a more traditional heavy-industry operation. If it comes back online at full capacity, it could employ well over a thousand workers directly, and a larger multiple indirectly through downstream supply chains. That is where the labour density in these three MOUs is genuinely located.

Which means the "5,000 jobs" figure is not, as it is being presented, a data centre number. It is an aggregate across three very different industries, dominated by the steel plant's labour density and padded by construction-phase jobs across all three projects. If the country asks the specific question – how many permanent jobs will the two data centres, at 450MW combined capacity, actually produce? – the honest answer is likely in the low hundreds at most. That is not nothing. But it is not 5,000, and it is not what the Government's rhetoric implies.

The country deserves the specific breakdown. Foreign and CARICOM Affairs Minister Sean Sobers, who signed the MOUs, should be asked to provide it. So should the Prime Minister.

The community that will host the first special economic zone has also been asking questions that the Government has not answered. CNC3 reported that Padarath did not respond to calls from journalists seeking comment on the concerns of Debe residents themselves, who have wanted to know what an AI data centre actually is, whether the promised jobs will reach them, and whether their water will be secure. That is a small but revealing gap. The country's largest new industrial commitment is being agreed at the top and defended in the abstract, without meaningful engagement with the citizens who will live nearest to it.

The extraction, in plain terms

Let me now put the central argument as directly as I can.

Data centres in Trinidad and Tobago will require the following:

• Electricity from the national grid, at rates negotiated for large industrial customers, subsidised where appropriate by the country's gas reserves.
• Water for cooling, drawn from either the national supply or from purpose-built desalination facilities that citizens will ultimately fund, through their water bills or through the Treasury.
• Land, at Picton Estate near Debe, or at the old sugar factory at Usine Ste Madeleine, or at Brechin Castle, or at Point Lisas, or at the site near the Beetham Water Treatment Plant that the Prime Minister has flagged.
• Tax incentives, likely in the form of holidays or reduced corporation tax rates for the operating period.
• Regulatory forbearance, in the form of expedited environmental approvals and streamlined permitting.
• Local labour for construction, and a smaller pool of local labour for operation.

The country will receive:

• Some corporation tax revenue, once the tax holidays expire. Some property tax revenue.
• Some construction-phase employment.
• A smaller number of permanent operational jobs, mostly filled by expat specialists in the first years unless a serious skills-pipeline commitment is made.
• Some indirect economic activity from the presence of the facilities.
• And, in the Prime Minister's phrasing, "future prosperity" in the abstract.

The country will not receive:

• Any share of the AI computation revenue generated by the facilities.
• Any equity in the operating businesses.
• Any control over what workloads are run on the machines built on our soil.
• Any right to route the data flowing through the facilities.
• Any technological capacity that transfers back to Trinidadian institutions.
• Any measurable increase in domestic AI or software capability.
• Any of the strategic optionality that comes from owning, rather than hosting, computational infrastructure.

This is the shape of the deal on the current disclosure. It is possible that fuller disclosure will show equity arrangements, technology transfer commitments, local training pipelines, or data sovereignty regulations that would change the picture. The country has not been told. The MOUs are non-binding. And the pattern of how deals of this shape have historically worked, in this country, is not encouraging.

The value chain of AI runs from raw compute – electricity and water and silicon – through training and inference and deployment, to end customers. The country is being invited to occupy the raw compute end of the chain. That is the point at which value is generated for whoever operates further up. It is not the point at which value is captured by whoever hosts.

If the argument sounds familiar, it is because it is. The oil and gas industry followed the same structure for eighty years. The bauxite trade did the same across the wider Caribbean. Sugar did the same before that. What is being proposed now, in a new industry, at a strategic moment in the global compute race, is a version of the same arrangement. The country hosts. The extraction goes elsewhere.

The signals from the signing

There is one further observation worth making, because it tells us how the Government is choosing to think about these deals.

The MOUs were signed by Sean Sobers, the Minister of Foreign and CARICOM Affairs. Not by the Minister of Public Utilities, Barry Padarath. Not by the Minister of Energy and Energy Industries, Roodal Moonilal. Not by a Minister for Digital Transformation, which the current administration has not created. The signature came from Foreign Affairs.

That signature tells us something. Data centres are being treated as foreign policy assets rather than as domestic infrastructure decisions. The venue of the signing – the US Embassy's Independence Day event – tells us the same. The choice of US firms specifically, as opposed to any global data centre operator from Europe or Asia or the Middle East, tells us the same again. These deals are part of the wider US-alignment arc that has been running through this Government's foreign policy since May 2025. That arc includes the Rubio meeting in St Kitts in February, the country's position on Venezuela, and the tensions with the CARICOM Secretary-General. Padarath has since been brought out to defend the water and energy specifics, promising man-made ponds at Debe in the short term and desalination plants at Moruga, Tableland and Mayaro in the longer term. He was not, however, part of the signing. His ministry inherited the utility implications of an agreement it did not negotiate.

None of that is illegitimate. Countries make foreign policy choices. Alignments have costs and benefits. A closer relationship with the United States may prove genuinely advantageous for Trinidad and Tobago's economic diversification. What is worth naming is that these data centre deals are not being negotiated as economic infrastructure by ministries that would ordinarily hold that responsibility. They are being negotiated as diplomatic instruments by the Ministry of Foreign Affairs. That is a signal about intent, and reasonable citizens should read it.

What actual technology development would look like

There is a version of a data centre deal that would land in a different place. It would include, in the MOUs themselves rather than as future promises, specific commitments to a local training pipeline that would fill the operational jobs with Trinidadian workers within three to five years. It would include equity participation for the country in the operating businesses, not just corporation tax on their profits. It would include data sovereignty regulations that would limit what workloads could be run on facilities on Trinidadian soil, so that the country retained some strategic optionality over what its infrastructure was being used for. It would include a serious investment fund, matched by the operators, for domestic AI startups and technology firms. It would include technology transfer arrangements that would build capacity at UWI, at the University of Trinidad and Tobago, and at the country's technical institutes.

That version of the deal exists in the world. Ireland has extracted some of it from its data centre operators. Singapore has extracted more. Even smaller states have negotiated harder terms than the current MOUs disclose. Whether Trinidad and Tobago can extract the same is a function of its negotiating position, its policy sophistication, and the political will of the Ministry of Foreign Affairs to insist on terms that go beyond hosting.

The country deserves to know whether that negotiation is happening. So far, on the public record, there is no indication that it is.

The choice being made

The Prime Minister has defended the deals with two arguments. The first is that any project will have environmental impact and that planners will cater for it. The second is that citizens who signed the petition are hypocrites, because Trinidad and Tobago is already environmentally degraded through litter and dumping. On the first, she is entirely correct. Environmental impact is a technical matter for engineers and regulators. On the second, she is wrong in a way that matters. Citizens raising concerns about a large new industrial commitment do not lose their standing because the country has other environmental problems. If anything, the presence of the existing problems is a reason to be more careful about layering new ones on top, not a reason to dismiss the people raising the alarm. Governments answer to their citizens. That is what governments are for.

Below the environmental question, though, sits the harder one. What is Trinidad and Tobago choosing to be? The country can be a hosting utility for foreign AI operators, contributing raw compute and receiving royalties, tax revenue, and a small number of specialist jobs. Or it can be a technology economy in its own right, with policy, training, and ownership structures that keep some of the value chain inside the country. The current MOUs, on the disclosure available, point strongly in the first direction.

That is a choice. It may turn out to be the right one. The country has run this model before, with mixed but genuine results. There is a case for a diversified extractive economy in a small state that lacks the scale to compete on the value-added end of most industries. There is also a case, made across the Global South by policymakers who have watched the same deals unfold in their own countries, that this particular moment in AI infrastructure is different, because the returns to compute ownership are compounding at a rate that leaves hosting economies further behind with every year that passes.

Both cases deserve a serious hearing. What the country has had, so far, is neither. It has had headline announcements, environmental critique, and rhetorical defence. The version of the argument which asks whether the specific deals now on the table capture enough of the AI value chain to justify the environmental and infrastructure costs, has not been made anywhere I have seen. That is the discussion the country should have before the concrete is poured.

Data centres are physical infrastructure. Once they are built, they are built. The negotiating window is now. The disclosure the country deserves – of financing structures, tax arrangements, training pipelines, equity terms, and data sovereignty commitments – is available only if it is demanded. And the ministry negotiating the deals is one whose remit is not usually where economic infrastructure lives. The Prime Minister has staked the Government's economic credibility on these deals. She should be asked to defend them, in specifics, before the country writes the cheque it will still be paying in 2050.

The refinery is back. It is a different shape. The country would be right to make sure it is a different deal.