Large data centers are no longer just digital infrastructures. They are starting to resemble energy-intensive industries with their own strategic energy approaches. In the United States, hyperscalers have crossed a line that a few years ago would have seemed exaggerated: they are no longer just buying green electricity or signing PPAs, but actively participating in new generation, grid, storage, and even nuclear projects.
Microsoft signed a 20-year power purchase agreement (PPA) with Constellation that will facilitate the reactivation of the Three Mile Island plant, renamed Crane Clean Energy Center, with 835 MW of carbon-free power expected by the end of this decade. Google has entered into the world’s first corporate agreement to purchase electricity from a fleet of small modular reactors, targeting up to 500 MW by 2035. Meanwhile, Meta is linked to a massive energy deployment in Louisiana to power its Hyperion AI campus, with new gas plants, transmission networks, storage, and associated renewable projects.
The message is clear: artificial intelligence is turning electricity into a competitive advantage. It’s no longer enough to have GPUs, land, fiber, and talent. What’s needed is firm, continuous power available 8,760 hours a year. And this necessity raises an uncomfortable question for Spain: is the regulation, the grid, and public debate prepared for large digital consumers to become leading energy actors?
Green PPAs don’t solve everything
For years, the dominant narrative has been simple: data centers will be powered by renewables. Solar and wind PPAs, origin guarantees, self-consumption, batteries, and commitments to climate neutrality. All of this matters and will remain part of the solution. Spain, moreover, has a very strong renewable position: Red Eléctrica projected renewable generation to account for 55.5% of the national electricity in 2025, rising to 56.6% when including self-consumption estimates.
But one thing is offsetting annual energy consumption, another is guaranteeing physical power every hour. An AI data center can’t shut down because there’s no wind in a January dawn or because solar production drops at sunset. Inference, training, storage, grid, and cooling loads require continuity. A 99.99% availability isn’t just a matter of accounting promises.
A renewable PPA can secure price, origin, and funding for new green capacity. But it does not guarantee that the specific kilowatt is available at all times at the grid node where it is needed. For that, backup, storage, flexibility, grid, and dispatchable generation are required.
| Energy solution | Advantage | Main limitation |
|---|---|---|
| Solar or wind PPA | Stable price and annual renewable energy | Does not guarantee constant hourly supply |
| Batteries | Flexibility, arbitrage, short-term backup | Limited duration and high cost at scale |
| Pumped hydro | Long-life storage | Location-dependent, permits, and availability constraints |
| Combined cycle gas | Firm, dispatchable, fast-start power | Emissions and reliance on gas supply |
| Nuclear | Low-carbon, reliable generation | Long timelines, complex regulation, social debate |
| Enhanced grid infrastructure | Resource sharing, resilience | Requires years of planning and investment |
The real debate shouldn’t be about renewables versus gas or nuclear as ideological blocks. The technical question is: what combination allows powering critical loads, reducing emissions, avoiding grid tensions, and not unfairly shifting costs to other consumers?
Gas makes a comeback thanks to AI
In the short to medium term, the technology best suited to meet the urgency of data centers isn’t nuclear but combined cycle gas. It’s faster to build than a nuclear plant, offers manageable power, can start and modulate production flexibly, and fills gaps when renewable generation isn’t enough. It’s not the ideal climate solution but provides an immediate technical response to loads that cannot be interrupted.
This explains what’s happening in the U.S. Nuclear power appears as a long-term strategic bet, but gas is the practical tool for expanding capacity now. In Louisiana, Entergy has announced agreements linked to Meta to develop over 5.2 GW of additional gas plants, alongside already approved power generation, transmission, batteries, and other resources. The scale is industrial, not minor.
Spain isn’t in the same situation as the U.S. It has a different, more interconnected electricity system with European policies, a high renewable mix, and more restrictive regulation for new fossil fuel plants. But the physical challenge is the same: if data centers grow rapidly and concentrate demand at specific nodes, firm capacity becomes a bottleneck.
Red Eléctrica started publishing demand access capacity on the transmission network in 2026, a crucial measure because large loads—including data centers—compete for connection points and planning. The debate is no longer abstract: the available network capacity can decide where future digital campuses are built and which projects are shelved.
Spain has renewables, but needs guaranteed power
Spain starts with clear advantages. It has abundant solar resources, a solid wind base, available land in certain regions, established data center operators, projects underway in Madrid, Aragón, Castilla-La Mancha, Cataluña, and others, and an attractive geographic position for interconnections, submarine cables, and a European sovereign cloud.
It also faces tangible challenges. Renewable generation is plentiful but variable. Grids aren’t built overnight. Permits are slow. Substations and transmission lines require planning. Batteries help but can’t fully replace firm generation during prolonged episodes. And Spain’s nuclear timeline, if unchanged, reduces some of the stable available capacity over the next decade.
The PNIEC 2023-2030 envisions high renewable penetration and a 22.5 GW storage target by 2030. While aligned with the energy transition, the influx of massive digital loads may require revising the pace and location of investments. Renewable energy in the system isn’t enough; it must arrive where and when consumption occurs.
Madrid’s case exemplifies this issue. The region hosts connectivity hubs, enterprise clients, cloud centers, and demand for data centers, but isn’t a surplus renewable area. Aragón, Castilla-La Mancha, and Andalucia may offer better land-energy combinations but need infrastructure, fiber, permits, and clients willing to accept less central sites. AI could influence the distribution of data centers toward locations where energy is easier to secure.
Will hyperscalers build their own generation in Spain?
The question isn’t just “if they can build combined cycles.” In Spain, any generation project requires permits, environmental assessments, grid access, regulatory approval, and economic viability. A hyperscaler can’t just build a thermal plant like a generator set. But they can promote dedicated generation schemes, long-term contracts, agreements with utilities, hybrid projects with batteries, direct consumption, industrial self-generation, or behind-the-meter solutions—always within the legal framework.
It’s unlikely we’ll see a tech giant operating gas plants under its own brand. More probable are partnerships with utilities, energy developers, and infrastructure investors to ensure firm capacity linked to specific campuses. This approach is already being tested in the U.S.: the tech company may not become a utility but can finance, demand guarantee, or structure contracts that make new generation viable.
Spain will need to decide what it accepts and under what conditions. If a data center needs 200 MW, 500 MW, or more, someone must answer three questions: who funds the grid upgrades, who guarantees capacity, and who bears emissions or backup costs when renewables are insufficient. If the answer is “the system,” costs can be shared among consumers. If “the developer,” then contractual and technical mechanisms will be necessary to make it possible.
| Questions for Spain | Why it matters |
| Who pays for new lines and substations? | Prevents shifting private costs to the whole system |
| What firm capacity must a large data center demonstrate? | Reduces risks to supply security |
| How is hourly renewable consumption measured? | Prevents confusing annual offsetting with actual supply |
| What role will batteries, gas, and nuclear play? | Defines the energy architecture for AI |
| Where should new campuses be located? | Alleviates Madrid and brings data centers closer to generation |
| What efficiency and waste heat requirements will there be? | Improves territorial impact of projects |
The debate Spain has yet to have
The data center industry often emphasizes sustainability, efficiency, advanced cooling, water use, and renewable energy. While these are essential, AI introduces an additional layer: denser loads, faster growth, and near-permanent demand. An AI campus doesn’t behave like a traditional hosting or colocation data center with moderate densities.
The idea that everything will be resolved with green PPAs is comforting but incomplete. Relying solely on gas is equally insufficient. The system needs a more comprehensive combination: renewables, storage, grid, demand management, firm generation, territorial planning, and transparency about costs. It also must distinguish between annual energy purchased and actual available capacity.
Spain can turn this tension into an advantage if it anticipates. It has renewable resources, strong energy companies, a growing data center industry, and the potential to attract technological investment. But delaying the debate risks getting stuck at connection points, permits, local opposition, and supply doubts.
The U.S. has already begun responding in its way: long-term nuclear, short-term gas, direct generation financing, and growing partnerships between tech firms and utilities. Europe, and Spain in particular, must craft its own model, balancing decarbonization, supply security, and digital competitiveness.
At its core, the question isn’t whether we want data centers. They’re already part of the digital economy. The real question is: what energy system do we want to build around them? Because if AI needs firm electricity 8,760 hours a year, someone will have to generate, transmit, and pay for it.
Frequently Asked Questions
Why do data centers need firm energy?
Because they must operate continuously. Their compute, network, storage, and cooling loads can’t depend solely on sunshine or wind at any given moment.
Does a renewable PPA guarantee 24/7 electricity?
Not necessarily. A PPA can offset annual energy and finance renewables, but doesn’t alone ensure physical, hourly supply at the point of use.
Could hyperscalers build combined cycle plants in Spain?
They could promote dedicated generation or utility agreements, but any plant requires permits, environmental assessments, grid access, and regulatory approval. It’s not just a private decision.
What technologies can support AI?
Gas, nuclear, hydro, storage, strengthened grid, and demand response can all be part of the solution. The key is combining firm capacity with decarbonization goals.