Where the narrative ends

For months, that line felt less like a slogan and more like an inevitable conclusion.

AI was pushing data center projects toward scales that would have seemed exaggerated not long ago. Electricity had stopped being a hidden operating cost and become a strategic constraint. Gas, generation, transmission, land, water, fiber and construction time were starting to show up in conversations that used to revolve almost entirely around chips and models.

And yes, Venezuela seemed to fit that story perfectly.

Venezuela has oil. It has gas. It has an enormous hydroelectric infrastructure. It sits close to the Caribbean, the Atlantic and the Gulf of the United States. And it genuinely needs to rebuild infrastructure, recover industry and find new ways to monetize assets that have been essentially idle.

The work we were doing at Frontier Grid seemed to be pointing there. For months we tracked large load requests, generation projects, regulatory changes, turbine demand, transformers, pipelines under construction, new contracts and permits. We weren't trying to confirm that AI would need more energy. That pattern was already too obvious.

The question became something else: where was it actually possible to find that energy in time?

We expected to find a new advantage in resource-rich geographies, like Venezuela. But what ended up happening is that we started discovering how much the word energy was hiding.

A requested megawatt is not necessarily an approved megawatt. An approved one may not be financeable. A financeable one may not have equipment. A generated one may not be able to connect to the grid. And even a deliverable one may not have the conditions to sustain a variable load for years.

Every signal was making the global thesis stronger. And also making it harder to apply to Venezuela.

Global scale becomes local pressure

The growth of data centers is real, but its impact tends to get described in a way that confuses global scale with local pressure.

The International Energy Agency estimates that data centers consumed around 485 TWh in 2025 and could approach 950 TWh by 2030. That would be roughly 3% of global electricity demand: an enormous expansion for a single sector, but not a total absorption of available energy worldwide. The real problem is that demand doesn't appear evenly distributed. It concentrates in a few places, arrives in large blocks and needs to connect far faster than traditional electrical infrastructure tends to get built.

A data center can go up in two or three years. A new power plant, a transmission line or a significant system expansion can take much longer, sometimes decades. It's a math problem: the speed of processing and the speed of physical infrastructure don't match.

In the United States we started seeing that gap in something close to real time. NERC raised its peak demand growth projection for the next decade to 224 GW, 69% higher than its previous assessment. New data centers represent the largest part of the expected increase. But that number doesn't mean 224 GW of projects will necessarily get built. And it reflects another problem: grid operators are still trying to figure out how much demand is real, how much is duplicated and how many projects will actually survive to operation.

That uncertainty is already changing the rules. Ohio just created a specific rate category for data centers, with long-term commitments and minimum demand charges. FERC is developing rules for loads above 20 MW and reviewing how to connect large consumers directly, without transferring unnecessary risk to the rest of the system.

This points to an important tension. The United States isn't just facing a potential electricity shortage. It's also facing a shortage of certainty.

Utilities need to build for a load that doesn't exist yet. Developers want to reserve capacity before committing all their capital. Regulators are trying to prevent existing households and industries from ending up paying for infrastructure that supports projects that may never materialize.

In the end, the AI energy problem turns out to be a financial problem too. Who guarantees that demand? Who pays if the data center's generation gets delayed? What happens when multiple companies request capacity for the same project across different jurisdictions?

Not even gas is a shortcut

As we kept following those signals, the original thesis started to shift.

The advantage doesn't simply belong to whoever has the most gas, the most sun or the most installed capacity. It belongs to whoever can turn a resource into firm, financed, contracted and deliverable power within a specific commercial window.

And that difference is enormous. Not even gas is a shortcut.

Natural gas came back quickly into the AI conversation because it can offer firm generation and, under certain conditions, can be installed close to the load. But even that solution has its own constraints.

The IEA estimates that between 15 and 27 GW of onsite gas generation could feed data centers in the United States by 2030. However, to sustain critical and variable loads, those systems will need between 30% and 70% more generation capacity than the site's nominal demand. The reason is straightforward: they have to be able to survive maintenance, failures, load variations and unit losses.

A 1 GW campus doesn't necessarily need to buy 1 GW of equipment. It may need considerably more. And turbines are entering a new wave of demand. Global orders increased 70% from last year. Manufacturers are accumulating years of backlog.

In other words: the supposed shortcut also needs equipment, fuel, compression, redundancy, permits, interconnection, contracts and capital. And that's precisely why the geopolitics of AI isn't just playing out in speeches, tech alliances or chip controls. The energy game lives in the turbine factory's order book, in the availability of a transformer, in the capacity of a pipeline, in a utility's balance sheet and in the bank that decides whether a transaction can actually be completed.

Venezuela looks extraordinary on paper

All of that is what brought us back to the case of Venezuela. Not because the answer is obvious, but because in that context Venezuela has certain elements working in its favor and others that aren't, and they're not always obvious at first glance.

On paper, the case for Venezuela's energy potential looks extraordinary. Venezuela doesn't just have the largest proven oil reserves on the planet. It also has significant gas resources and a hydroelectric system built around continental-scale assets. Its location connects it to the Caribbean, Trinidad and Tobago, the Atlantic, Guyana and Suriname, and puts it relatively close to the industrial system of the American Gulf.

The problem is that this collection of gas projects across northern South America is not an integrated network. At least not right now. Each project is different, operating under different jurisdictions, competing for capital, buyers, equipment and political attention.

The new agreement to rehabilitate Tocoma and Macagua is a useful example of what that process actually looks like in practice. IMPSA is proposing to bring in 160 MW initially through two Macagua units, recover 672 MW between both complexes within 24 months and eventually reach up to 2,640 MW. There are identified units, contractors, timelines and a clear execution sequence.

But the announcement talks about the first 160 MW within 90 days of signing the contract, and also notes that work will begin once the administrative phase is complete and the notice to proceed has been issued.

So when does it actually start? In a news broadcast that difference might not seem important. Through the lens of infrastructure, it can change everything.

The project could meet its first milestones. It could also get delayed. It could run into additional problems when crews open up equipment that has been sitting idle for years. It could recover generation capacity without yet resolving the ability to move that electricity toward the main consumption centers.

Those 160 MW matter. But they don't yet prove a systemic recovery.

We also don't have complete enough public data to verify some of the claims circulating about electricity potentially stranded in Guayana. It's possible that additional generation exists that can't be fully distributed across the rest of the country. It's also possible that effective capacity is lower once you account for unit availability, hydrological seasonality, stability, maintenance and contingencies.

The truth is we haven't found data that lets us answer that question with confidence.

And yes, that's frustrating. It's also part of what working in infrastructure actually looks like.

In this kind of work, some signals come with a number that seems significant but no clear methodology. Others have an operator but no published financing. A project can have a resource and a plan but no final investment decision. A project can have a date but no clear definition of when the clock actually starts.

The temptation is to fill those gaps with optimism or with fatalism. Both are ways of inventing certainty.

A resource is not a project

Venezuelan gas offers a more flexible path than waiting for the complete reconstruction of the national grid. It could be used for dedicated generation, industry, petrochemicals, critical infrastructure or energy nodes partially isolated from the weaker system. It could also flow toward existing facilities in the Caribbean and the Atlantic.

The signals of that are becoming more visible. Shell still classifies Dragon and Loran as important pre-FID options, tied to existing infrastructure in Trinidad and Tobago.

The gas exists. The fields are identified. The operators and receiving infrastructure are there. But commercial decisions, licenses, contracts, financing, payments, equipment and execution still remain. That's precisely the difference between a resource and a project.

The EIA has noted for years the limitations of Venezuela's infrastructure to process, store, distribute and fully use its gas. Having enormous reserves doesn't eliminate the need for wells, gathering, compression, processing and pipelines.

And once that molecule is available, another unresolved question appears: who ends up with it?

It can be monetized outside Venezuela. It can feed domestic generation. It can support an industrial recovery. It can back ports, hospitals, telecommunications and other critical assets. It can supply private nodes with their own contracts. And eventually it could feed compute.

Those paths are not equivalent. And they can't all happen at the same time.

AI may raise the strategic value of Venezuelan gas, but it doesn't have to become its first buyer or its most urgent use. In fact, one of the conclusions that has taken us the longest to accept is that Venezuela could benefit from the energy reorganization driven by AI without becoming a major AI hub.

It can sell gas. It can recover generation. It can attract capital toward certain assets. It can partially integrate into a much larger Atlantic energy system. It can rehabilitate strategic infrastructure long before it resolves its national electricity system.

That alone would be a meaningful transformation.

Recovery will not be national, uniform or immediate

The media conversation tends to treat politics like a light switch. There are elections and reconstruction begins. There are no elections and nothing can change.

Infrastructure rarely works that way.

A political change could expand access to capital, improve contractual conditions, modify sanctions and attract operators. It could accelerate projects that aren't financeable today. It wouldn't automatically repair transmission. It wouldn't manufacture turbines faster. It wouldn't turn a gas reserve into production. It wouldn't build in months the institutional capacity needed to simultaneously execute dozens of complex projects.

Political continuity doesn't mean absolute stillness either. Under imperfect conditions, licenses can advance, rehabilitations can happen, specific agreements can be reached and projects can move forward through carefully structured commercial arrangements.

The cost may be an uneven recovery: reliable energy nodes surrounded by a still-fragile national system; production for export while internal deficits persist; greater dependence on foreign operators and capital; strategic assets advancing faster than basic public services.

None of that constitutes an orderly transition. It doesn't mean the absence of progress either.

Over the next several years, Venezuela can advance, retreat and open opportunities at the same time. One project can work while another stalls. A license can unlock capital and then get revised. Generation can come back without the ability to repair the full grid. Gas can reach the market while domestic industry remains constrained.

That scenario is less satisfying than a national reconstruction narrative. It's probably more plausible.

We could still be wrong in either direction.

Maybe the combination of external capital, modular generation and new contracts will allow energy nodes to develop much faster than currently seems possible. Maybe certain projects can insulate themselves from national weaknesses and operate with enough reliability to attract industry or compute capacity.

The opposite could also happen. Early announcements may not survive the administrative phases. Accumulated deterioration may be worse than estimated. Financial constraints may prevent maintenance and spare parts. Gas may find better buyers outside the country. AI demand may grow more slowly, shift to other regions or need less energy per unit of work.

We don't have that ambiguity resolved. And we don't want to resolve it before the evidence exists.

Our position isn't that Venezuela lacks an energy opportunity. It's that the opportunity doesn't appear simply because a global narrative needs a new protagonist.

Where the narrative ends

We started out looking for where artificial intelligence would find its energy. We ended up finding a harder question:

Where can an energy resource become reliable capacity before the market, the technology, the government or the cost of capital changes?

For Venezuela, the answer probably won't be national, uniform or immediate. My suspicion is that it will be specific.

A gas field. A rehabilitated power plant. A transmission line with available capacity. A port. An industrial zone. A contract that can actually be collected. An operator that can maintain the asset. A node capable of surviving even as the system around it remains weak.

AI can add pressure. Gas can open a path. Politics can expand or close that space. But none of those things replace the physical sequence.

Venezuela doesn't need to become an AI powerhouse to recover strategic relevance. It needs projects capable of surviving the distance between the announcement and the operation.

That distance is where the narrative ends. And where the infrastructure begins.