Data centres and AI: what the new electricity demand means for solar and storage investors

Every revenue calculation for a solar park or battery storage project rests on an assumption about future electricity prices, and prices come out of the balance between supply and demand. The supply side gets plenty of attention: build-out, auctions, market premium. The demand side usually appears in project documents as a single sentence pointing to data centres and artificial intelligence. This article examines what sits behind that sentence: what data centres consume today, what has been registered in grid connection procedures, what the Energy Efficiency Act requires of them from 2027, and how much of it actually reaches an investor.

How much electricity do data centres in Germany actually use?

Around 21 billion kilowatt hours in 2025. That figure comes from the study the Borderstep Institute produces annually for the German digital association Bitkom; the year before it was 20 billion, and in 2015 it was 12. Consumption has therefore almost doubled within a decade, and it continues to grow.

What matters is the yardstick you hold it against. Germany's gross electricity consumption in 2025 was 526 terawatt hours according to the Federal Environment Agency. Data centres therefore account for roughly four percent of German electricity consumption. That is a noticeable block, but it is not the item that moves the power market today; industry consumes a multiple of it.

How fast is demand growing, and how much of it is AI?

Measured by connected load, considerably faster than by consumption. The installed connected load of German data centres rose by nine percent in 2025 to 2,980 megawatts; by 2030 the study expects more than 5,000 megawatts, close to a doubling against 2024. The growth is regionally concentrated: Hesse accounts for over 1,100 megawatts, Bavaria for 420 and North Rhine-Westphalia for 378.

The AI share is where the real movement sits. Data centres built specifically for AI workloads currently reach 530 megawatts, around 15 percent of connected load. By 2030 that is expected to be 2,020 megawatts and 40 percent. Operators are backing this with capital: in 2025 alone, roughly 12 billion euros went into IT hardware and a further 3.5 billion euros into buildings and building services.

Why do the grid operators quote far larger numbers?

Because they are counting something else. For the Grid Development Plan for Electricity 2037 with an outlook to 2045, the Federal Network Agency approved a scenario framework in which data centre electricity consumption is reported separately. Depending on the scenario the values are 78.4, 97.3 or 116.2 terawatt hours, and they are identical for 2037 and 2045 because every project included is already planned for 2037. To convert capacity into consumption the agency assumes 5,000 full load hours.

Working backwards, 78 to 116 terawatt hours correspond to a connected load of roughly 16 to 23 gigawatts. The industry study expects a good 5 gigawatts for 2030. The gap between the two is a factor of three to four and a half, and it is not an arithmetic error: the scenario framework draws on the transmission system operators' survey of large consumers, that is, on registered projects at different stages of maturity. The agency counts projects up to the status of „advanced planning“ in full, and those merely at „planning“ status at 25 or 50 percent depending on the scenario.

What is striking is how clearly the Federal Network Agency qualifies its own numbers. The updated survey raised the reported demand by around another 50 percent; whether all projects will be built at the registered connection capacities „can be doubted“. Registering in a grid connection procedure is cheap and non-binding, and only part of it gets built. Anyone who knows this dynamic from the storage side will recognise it: there, too, registered capacity far exceeds what eventually reaches the grid (Germany's battery storage market: the numbers, the drivers, and why "now" has a date).

For investors this is the single most important distinction in the whole topic. A queue is not a forecast. When a project document argues with double-digit gigawatt figures for data centres, it is relying on registrations, not on plants.

Why must data centres source renewable electricity from 2027?

Because the Energy Efficiency Act requires it. §11 (5) EnEfG obliges data centre operators to cover their electricity consumption on a balance-sheet basis 50 percent from renewable energy from 1 January 2024, and 100 percent from 1 January 2027. This is not a statement of intent or a forecast, but applicable law with a fixed date. Under §3 no. 24 EnEfG a data centre is any structure from 300 kilowatts of non-redundant rated connected electrical load; pure network nodes are excluded.

The provision does not stand alone. Data centres entering service from 1 July 2026 must achieve power usage effectiveness of no more than 1.2 under §11 (2) EnEfG and reuse at least 10 percent of their energy, rising to 15 percent from July 2027 and 20 percent from July 2028. Older facilities must reach 1.5 from July 2027 and 1.3 from July 2030 under subsection 1. The legislator is therefore demanding both less consumption per unit of computing and clean sourcing for whatever remains.

This is where one word deserves attention. The obligation is to cover consumption „on a balance-sheet basis“, and that means through certificates, not necessarily through a contract with a particular plant. Guarantees of origin are tradable across Europe and comparatively cheap to obtain. A long-term power purchase agreement with a German solar park is one way to comply, and an attractive one for many operators because it also fixes their purchase price for years. It is not the only way. How such contracts are built and what their reliability depends on is set out in Understanding PPAs: how power purchase agreements make solar park revenues predictable.

What does this mean for the market value of a solar park?

Less than the headlines suggest, and the reason lies in the load profile. A data centre runs evenly; in its scenario framework the Federal Network Agency assumes 5,000 full load hours a year, spread across day and night, summer and winter. A solar park generates in a concentrated, weather-dependent way, peaking around midday. Additional steady demand mainly lifts prices in the hours when photovoltaics deliver little anyway.

The problem that depresses the market value of solar power is a different one: oversupply during its own generating hours. In 2025, according to Federal Network Agency figures, wholesale prices were negative in 573 of 8,760 hours, against an average day-ahead price of 89.32 euros per megawatt hour. A load spread evenly across the day does not clear those midday hours. What happens economically in them and which rules apply is covered in Negative electricity prices: what they mean for solar and storage investors; how a solar park's revenue is composed in the first place is shown in Direct marketing and the market premium: how a solar park earns its money.

Where the demand does arrive is on the offtake side. An operator who has to organise demonstrable renewable sourcing by 2027 and at the same time wants planning certainty over ten or fifteen years is a plausible counterparty for long-term offtake agreements. That works through the contractual landscape rather than the exchange price, and it reaches an individual project only if that project actually holds such a contract.

Does battery storage benefit from the data centre boom?

In two directions that run against each other. A storage asset earns on the difference between cheap and expensive hours. Additional steady base load tends to lift prices in the cheap hours during which a storage asset charges, which narrows the spread. At the same time, growing load in consumption-heavy regions meets a grid that is not expanding at the same pace, and regional bottlenecks raise the value of flexibility in the right place.

Which of the two effects dominates depends on location, grid situation and actual build-out, and it cannot be quantified credibly today. A provider who cites the data centre boom as an argument for wider spreads should therefore be able to explain which of the two directions is assumed and why. What a storage asset's revenue is made of is shown in Direct marketing explained: day-ahead, intraday and balancing power; the orders of magnitude involved in Battery storage returns: where the revenue comes from, and what is realistic.

Where does the competition for grid connection arise?

At the same points energy projects compete for. In the scenario framework the Federal Network Agency notes that the majority of registered data centre projects are to be connected in the distribution grids, which are precisely the voltage levels ground-mounted solar and battery storage go to as well. Regional concentration sharpens this: over a thousand megawatts of registered data centre load in Hesse meet a grid that was not designed for it. In the consultation on the scenario framework it was explicitly doubted that the Frankfurt area could support a development towards 5,000 megawatts.

For a direct investment this is not an abstract market question but a due diligence question about the specific project: connection capacity is the scarce good, which makes a solid grid connection commitment one of the most valuable documents in the data room. How to judge its reliability is described in Grid connection, easement, building rights: how to tell whether a direct investment has truly secured its site; what costs attach to the connection and how two assets can share a connection point is shown in Grid fees for battery storage: the exemption until 2029 and the AgNeS reform and Co-located vs. stand-alone: which gives the better risk structure.

What follows for a direct investment, and what does not?

Three things are demonstrable: data centre demand is growing, it is growing above average in the AI segment, and from 2027 it must by law be covered 100 percent from renewable energy on a balance-sheet basis. That is structural support for the asset class as a whole, and it rests on a statute with a date rather than on a market opinion.

What is not demonstrable is the step from there to a specific return. A demand trend is not a price commitment; it says nothing about what a particular solar park will earn per megawatt hour in a particular year. A profitability calculation that justifies higher electricity price assumptions with the AI boom has a story at that point, not an argument. What carries a project are contractually secured revenues and conservative assumptions for the rest; which levers come together is shown in Investing in solar parks: revenues, costs and tax leverage at a glance and Investing in battery storage: the options at a glance.

Questions to ask the provider

1. Which electricity price assumption underlies the revenue calculation, and what is its source? A demand narrative is not a source.
2. Is there a long-term offtake agreement, and if so: who is the counterparty, over what term and at what price?
3. If data centres are cited as an argument: does the figure refer to plants actually built or to registered grid connection requests?
4. For storage: does the calculation assume that additional base load widens or narrows spreads, and on what basis?
5. How reliable is the grid connection commitment, and in which grid region does the project sit?

Which electricity price assumptions a specific project calculation rests on, and whether the revenue side is contractually underpinned, is something we work through against the actual documents in a no-obligation initial conversation, with assumptions named. We give no return guarantees.