With only a few days to go before the Móra la Nova artificial intelligence gigafactory bid is submitted in Brussels, there is a lot of talk about chips, about jobs and, rightly, about water. There is less talk about energy, which is the question that will decide the bid's credibility. Where will the electricity come from at three in the morning on a windless Tuesday in January?
An AI data centre is unlike any other consumer. It runs 24 hours a day, every day of the year, with almost flat demand. The 54 MW guaranteed for the first phase, running at full output all year, would mean up to around 470 GWh a year. The bid commits to 24-hour renewable coverage, and that is far more demanding than buying green certificates for the year as a whole. It means that every hour of consumption must be covered, at that moment, by renewable energy that is either being produced or has been stored.
No technology solves this on its own. Supply of this kind is built with a portfolio that combines solar for the sunny hours, wind, which often blows when the sun has gone, and storage to bridge the two. I have spent much of my career in molten-salt solar thermal plants, which store heat for ten hours or more and generate at night, and today cheap photovoltaics combined with batteries make it possible to design supply profiles that were unthinkable ten years ago. The grid remains the last resort for the hours the portfolio cannot cover.
The contract is as important as the technology. These plants are financed with long-term power purchase agreements, which normally fix an annual volume or a share of the plant's output as it is produced. On its own, none of these contracts ensures that any specific hour is covered. Hour-by-hour coverage comes from the portfolio, combining solar, wind and hybrid contracts with storage so that the combined profile comes close to the consumption profile, with an electricity retailer covering the rest. Designing it this way costs more than signing a conventional solar contract, but without this work, 24-hour renewables remain a slogan.
The flexibility of the data centre itself also helps. Training a model can be done in the hours when there is surplus energy, while user queries have to be served instantly. A centre that manages its demand with energy in mind needs less storage and pays less for its electricity.
Batteries have an essential role, and it is worth knowing exactly what it is. They shift energy from some hours to others. They store the midday solar surplus, when Spain already has many hours with zero or even negative prices, and return it in the evening and at night. The systems being installed today last between two and eight hours and are excellent for covering the evening, stabilising the grid and relieving congested substations. We see this in Chile, where we are developing a storage project in Atacama, one of the regions with the highest solar radiation on the planet. In 2025 some 6 TWh of solar energy was lost there because the grid could not absorb it, which is why batteries are growing so fast there.
That said, a battery does not work miracles. Covering a single winter night, some fourteen hours without sun, at the first phase's consumption would require around 750 MWh of storage. That is more than the entire 500 MWh platform we have in development in Atacama, and the following day several hundred megawatts of solar panels would be needed to recharge it. A week of cloud and calm, or storing summer energy for the winter, is beyond the reach of any battery available today. On the positive side, financing batteries will become easier, because the capacity market approved this September will allow new storage to secure contracts of up to fifteen years.
In Móra la Nova, the first phase has the advantage that the 54 MW are already guaranteed. The second phase, according to the mayor, needs about 75 MW more and two new lines from the Els Aubals substation, and will depend on new wind and solar generation and on storage. The new electricity grid planning, which is set to add up to 3,800 MW of demand capacity in Catalonia, will help. Even so, there are three decisions that would be worth taking now. The first is to design the hour-by-hour contract portfolio from day one, even if the first phase can run without new renewables. The second is to plan storage close to the load or at the substation itself, because grid capacity is the scarcest resource in the system and batteries multiply its value. And the third is to design the centre so that part of the load is flexible.
I would add one thing that costs little: that the gigafactory publish each year the percentage of the year's 8,760 hours it has genuinely covered with renewable energy. It would be the best answer to those calling for a public debate based on data, and it would turn the 24-hour promise into a figure that anyone could check.
In the projects we are developing in Nevada and Atacama, we start from this same logic. The Ribera d'Ebre has been generating energy for the whole country for decades, and if the gigafactory comes, it could also become one of the first places in Europe where artificial intelligence runs on renewable energy at any hour. Batteries will be essential, but without good contracts and a well-planned grid they will not be enough. Everything has to be thought through for three in the morning.