Artificial Intelligence & the Energy Factor

Energy is becoming a strategic factor in the AI era. Between grid bottlenecks, sovereignty and sustainability, where value is created is crucial. How Germany is turning from an electricity problem into a shaper of intelligent AI infrastructure - and why the course must be set now.

The global hunger for computing power is growing exponentially. Large Language Models (LLMs), agent-based AI systems and cloud services are driving a power boom that poses completely new challenges for the energy system. According to the International Energy Agency (IEA), the global electricity consumption of data centers will double to around 945 TWh by 2030 - the equivalent of Japan's current electricity consumption. At the same time, 80 to 90 percent of European cloud infrastructure is in the hands of American hyperscalers. In the USA, the first data centers are already unable to connect to the grid due to a lack of grid connections and generation capacity; in 2025, investments by US companies Amazon and Google in local data centers were withdrawn due to a lack of grid connection commitments. Meanwhile, a shadow energy infrastructure is emerging in the USA and demand for gas-fired power plants is exploding.

Germany must make a decision: Either it plays an active role in shaping the future - as a location for sovereign, sustainable AI infrastructure - or it remains a permanent net importer of digital value creation. New European AI data centers show that a third way is possible: ambitious, efficient and European. This article highlights the four key dimensions that will determine the success of this path: sovereignty, flexibility, resilience and sustainability.

German data centers consumed around 21.3 TWh of electricity in 2025 - around 4 percent of total German consumption and four times as much as a large city like Munich (FfE, 2026). The proportion of large data centers (over 5 MW) has increased significantly in recent years and already accounts for 50 percent of installed capacity today. By 2037, the share could rise to 10 percent of German electricity consumption (BNetzA NEP 2025).

The growth driver is clear: artificial intelligence. While classic database queries require little energy, a query to a large language model consumes around 2.9 Wh according to the IEA - ten times more than a Google search. The share of AI data centers in total installed data center capacity is expected to increase from the current 15 percent to around 40 percent by 2030.

The biggest barrier to growth is not a lack of demand, but a lack of network capacity. Waiting times of up to seven years for new network connections in Germany are slowing down expansion considerably - in the USA, the waiting time is one to three years. Over 9 GW of new connection capacity for data centers has been registered with the Federal Network Agency, but network expansion follows a decade rhythm, while data center operators can build in just a few years. In Frankfurt, there is effectively a grid lock for new AI data centers until 2030.

The dependency is structural

A very large proportion of European cloud infrastructure is provided by US hyperscalers such as Microsoft Azure, Amazon AWS and Google Cloud. The Bertelsmann Stiftung's EuroStack initiative puts the annual outflow to foreign tech providers at 264 billion euros. In addition, the US Cloud Act theoretically gives American authorities access to all data stored on US hyperscalers - regardless of their physical location. The potential economic damage is difficult to quantify.

Germany, on the other hand, accounts for just 5 percent of global AI computing power, while the USA accounts for 70 percent. This discrepancy is not a technical one, but a strategic one: if you don't own the computing infrastructure, you can't sovereignly shape the AI layer above it.

Sovereign AI as a blueprint

First AI factories in Europe are a strong signal. Recently, Germany's AI computing power was increased by 50 percent in one fell swoop. The decisive factor here is the sovereign architecture: data in Germany, German law, European personnel, in-house training of the SOOFI model.

The conclusion for the site

Sustainable data centers in Germany are the prerequisite for LLMs and agent-based AI systems to be set up and operated in Germany at all. Without its own computing infrastructure, the agentic layer - i.e. AI systems that plan, decide and act independently - will remain permanently anchored in American data centers.

This is not only critical from a data protection perspective. It affects the competitiveness of entire industries: Energy supply, health, production, mobility. If you don't build your own infrastructure here, you will have to buy it in later at a high price. The EuroStack initiative is calling for 300 billion euros over ten years; the German coalition agreement of 2025 explicitly references EuroStack.

AI as both problem and solution

Data centers are not just consumers of electricity - they can actively contribute to grid stability. The FfE distinguishes between two flexibility categories: process-oriented flexibility (workload shifting) and periphery-oriented flexibility (UPS, cooling system, emergency power supply).

Delay-tolerant workloads such as AI model training can be postponed in times of low electricity prices or high renewable feed-ins. Google DeepMind has achieved energy savings of 40 percent in the cooling area through AI-supported cooling management. The cooling system can contribute to peak shaving by pre-cooling when electricity prices are low.

Sector coupling: The data center thinks for itself

Smart data centers of the future will no longer just consume electricity, but will be actively embedded in the energy system. Similar to smart home storage systems and dynamic tariffs, spot price-controlled energy management will enable data centers to link their consumption to EPEX spot prices. Delay-tolerant AI training jobs are started when the electricity price is at its lowest - typically when there is a high wind or solar feed-in. This triggers a win-win situation: lower operating costs for the data center operator and simultaneous grid relief in the event of a surplus.

Battery storage and e-fleets in combination

The UPS batteries of a data center for uninterruptible power supply (UPS) are technically suitable for participating in the balancing energy market - provided that fail-safety is a priority. Even more interesting is the idea of managing data centers, large battery storage systems and company car fleets as a virtual power plant network: A 20 MW data center with 4 MWh UPS capacity, combined with a connected battery storage system and bidirectionally chargeable company vehicles in the campus parking lot, results in a significant combined flexibility resource. AI-supported energy management system (EMS) can optimize all flows in real time.

Local heating networks as a building block for the heat transition

A 10 MW data center can mathematically supply over 5,000 single-family homes with heat (FfE, 2026). Modern data centers feed their waste heat into a district heating network for around homes and offices. According to the DENA report, 3 TWh of data center waste heat could already be used by end consumers in 2035. Early coordination with municipal heating network planners is crucial here: anyone who plans a data center without considering the surrounding local heating network is giving away money and added value in terms of climate policy.

Arvato Systems, for example, is specifically building on local heat supply for the campus at the company headquarters - the energy and sustainability management system green.screen monitors the energy consumption and optimizes it.

In the USA, grid reserves are eroding at an alarming rate: eight out of thirteen regional electricity areas are approaching critical limits. Capacity prices have risen by over 1,000 percent so far. Microsoft has canceled more than 200 MW of planned AI capacity due to grid bottlenecks. The response of many US companies: 47 known off-grid power plant projects, often powered by natural gas - with an operating life of 25 to 40 years, a devastating lock-in in terms of climate policy.

Germany is taking a different approach: the Energy Efficiency Act (EnEfG) requires new data centers to be 100% renewable energy (on balance) from 2027, sets PUE limits and requires the use of waste heat. Despite a planned amendment, which is intended to give large locations in particular more flexibility, the basic framework remains ambitious. Europe covers over 85 percent of its additional DC electricity requirements from renewables and nuclear energy - compared to over 40 percent gas and coal in the USA.

For the resilience of critical infrastructure, it is also crucial that data centers are not only dependent on individual network nodes. Redundant supply paths, local generation capacity (on-site generation), cold reserves and geographically distributed workload distribution are not a luxury, but an operational necessity.

The duel of concepts is more vivid than almost any other example in the energy debate. On the one hand: the Monarch Compute Campus in West Virginia, planned with a capacity for 1.5 million households - in a state with a population of just 1.8 million. Off-grid, powered by gas-fired power plants, with a planned lifetime well into the 2050s. On the other hand, new AI factories: retrofit instead of new construction, river instead of gas turbine, district heating instead of waste heat destruction.

The big open question remains the Jevons paradox: technological efficiency gains can be overcompensated by absolute volume growth. Only a consistent combination of efficiency incentives with absolute capacity limits or CO₂ budgets can curb this rebound effect.

For the energy industry and grid operators

• Accelerate the network connection process: The previous first-come-first-served principle favors immature requests. Allocating capacity according to maturity and system benefit is more efficient.
• Conceive data centers as flexibility partners: Demand response, spot price integration and load shifting open up considerable flexibility potential - provided that network operators and data center operators develop common rules of the game.
• Planning local heating networks at an early stage: municipal heating networks and data center locations belong in the same infrastructure planning. Those who plan today without including waste heat potential will end up with expensive supplements tomorrow.

For companies and industry

• Design your own AI infrastructure with confidence: Anyone using LLMs and agent-based AI systems for critical processes should plan their operation on European infrastructure - not only for data protection reasons, but also for availability, latency and regulatory control.
• Thinking battery storage, e-fleets and data centers together: The corporate campus of the future operates an integrated energy management system (EMS) that optimizes server waste heat, UPS batteries, company vehicle fleets and dynamic electricity prices as an overall system - controlled by AI.
• Invest in efficiency, not just capacity: Power Usage Effectiveness optimization, liquid cooling and AI-supported data centre management pay for themselves - especially with rising energy prices.

For politics and municipalities

• Harmonize the regulatory framework: The EnEfG sets the right incentives; the amendment should not soften the basic requirements, but make them flexible - especially for the waste heat obligation.
• Designate strategic data center locations: Municipalities that specifically direct data centers close to heat networks or to existing network nodes create added value for all stakeholders.
• Implement EuroStack consistently: The reference in the coalition agreement must be translated into concrete funding programs for European AI computing infrastructure.

Data centers are at the heart of a technological and geopolitical tectonics that is only just beginning. At the same time, they are the biggest source of new energy demand and the most promising lever for grid flexibility, heat transition and sector coupling. This paradox is not a contradiction - it is a design task.

Germany has the industrial and regulatory substance to take on this design task: as a location for sustainable, sovereign AI infrastructure, as a pioneer in the integration of data centers into municipal energy systems and as a driver of the European path - compared to the American off-grid model and Chinese state subsidies with excess capacity.

This article is based on publicly available studies and reports. As of February 2026.