Europe is racing to build its own AI computing power rather than relying on distant data centers, and the infrastructure race is intensifying as grid constraints force companies to explore alternative energy sources. Cerebras Systems announced plans to deploy 200 megawatts of artificial intelligence data center capacity across Europe by 2027, targeting Norway, Finland, and potentially France, while simultaneously, Siemens and FuelCell Energy signed a partnership to develop 100MW-plus fuel cell power systems specifically designed for data centers facing grid congestion.

Why Are European Companies Building AI Data Centers Locally?

The shift toward regional AI infrastructure reflects a fundamental change in how enterprises think about computing. European businesses increasingly need low-latency access to AI tools for interactive applications like customer service agents, coding assistants, trading analysis, and public sector systems. When computing resources are distant, scarce, or subject to geopolitical risk, these latency-sensitive workloads suffer.

Cerebras selected its European deployment locations based on several strategic factors:

• Energy Availability: Access to cheap or clean electricity supply, which is critical for power-hungry AI infrastructure
• Climate Conditions: Cooler climates in Norway and Finland reduce cooling costs, a major operational expense for data centers
• Grid Policies: Favorable regulatory environments that support rapid deployment and interconnection
• Data Residency: Regulatory pressure to keep sensitive data within European borders rather than routing it to US-based cloud providers

The company's hardware approach differs from traditional GPU clusters. Cerebras uses wafer-scale systems designed for high-throughput AI training and inference, offering an alternative to the NVIDIA-dominated infrastructure landscape that most enterprises currently standardize around.

What's Driving the Fuel Cell Partnership for Data Centers?

While Cerebras focuses on chip architecture and data center deployment, the real constraint emerging across the industry is power itself. FuelCell Energy's business development pipeline has surged 275 percent since February 2025, with data center customers accounting for the majority of new interest. This explosive growth reflects a critical problem: grid interconnection queues and transmission congestion have stretched deployment timelines for new data centers, forcing operators to seek behind-the-meter power solutions.

"Grid congestion and interconnection backlogs are the primary demand drivers," noted CEO Jason Few of FuelCell Energy.

Jason Few, CEO at FuelCell Energy

The Siemens and FuelCell Energy partnership addresses this gap by combining FuelCell Energy's modular carbonate fuel cell systems with Siemens' electrical balance-of-plant infrastructure. Siemens will design and supply switchgear, transformers, protection systems, and microgrid controls that integrate fuel cell generation with battery storage and medium-voltage equipment. FuelCell Energy's modular systems come in 1.25MW, 2.5MW, and 12.5MW configurations that can be combined for utility-scale installations exceeding 100MW.

Carbonate fuel cells run on natural gas and produce baseload power with lower air emissions than diesel or gas turbines. They also generate usable waste heat that can offset supplemental cooling loads, making them particularly attractive for data center operators seeking to maximize efficiency.

How to Evaluate the Viability of Alternative Data Center Power Solutions

• Manufacturing Capacity: FuelCell Energy is expanding its Torrington, Connecticut facility from 100MW to 350MW annually to meet demand, more than tripling current output. The company's backlog already exceeds present manufacturing capacity, signaling genuine market traction rather than speculative interest
• Binding Commitments: FuelCell Energy signed a binding 380MW capital equipment purchase agreement with Fit Energy USA in June 2026, structured in four phases with an initial 30MW commitment. This is the company's largest deal to date and demonstrates customer confidence in fuel cell technology for data centers
• Integration Partnerships: The Siemens MoU signals that established electrical infrastructure providers view fuel cells as commercially viable distributed generation, not just demonstration technology. This institutional validation matters more than pilot projects alone
• Grid Constraints as Demand Driver: Data center operators are exploring fuel cells specifically because grid interconnection delays have become unacceptable. This is a structural problem, not a temporary bottleneck, making alternative power solutions increasingly essential

The broader context matters here. Global data center power demand is projected to rise from 104 gigawatts in 2025 to 132 gigawatts in 2026 and 290 gigawatts by 2030. In the United States alone, data center power demand is expected to climb from 31 gigawatts in 2025 to 41 gigawatts in 2026 and 66 gigawatts in 2027, more than doubling in two years. US data centers' share of peak summer electricity demand is set to nearly double, from 4.1 percent in 2025 to 8.5 percent by 2027.

What Challenges Remain for European AI Infrastructure?

Despite the momentum, significant obstacles remain. Cerebras must navigate slow power access and complex permitting processes across multiple European countries. Europe's AI infrastructure demand is colliding with grid limitations, sustainability commitments, and national industrial policies that vary by region.

The company also faces competitive pressure from hyperscalers, sovereign cloud projects, and other operators attempting to convert megawatts into AI revenue before market conditions shift. Additionally, there's uncertainty around capacity allocation. The Cerebras announcement references OpenAI as a customer, but if OpenAI receives priority access to early deployments, it may limit available capacity for other enterprise customers.

For investors and enterprise buyers evaluating these infrastructure plays, the key distinction is between announced capacity and operational capacity. Contracted megawatts do not equal operational capacity, and operational capacity does not equate to solid enterprise adoption. Success requires signed power agreements, operational sites, utilization rates, and commitments from non-OpenAI customers.

The infrastructure buildout underway is enormous, but value capture is highly concentrated. A separate analysis of the AI infrastructure stack found that six major buyers (Microsoft, Amazon, Alphabet, Meta, Oracle, and CoreWeave) will deploy over 760 billion dollars of capital expenditure in 2026, yet the profit pools concentrate in just five chokepoints: chip design, foundry, lithography, memory, and networking. Assembly and rental companies capture almost none of the value despite growing revenue 106 to 346 percent year-over-year.

For European enterprises, the substantive changes lie in latency, procurement location, and energy independence. As AI workloads become more interactive and latency-sensitive, the ability to access computing power locally becomes a competitive advantage. The convergence of Cerebras' regional data center strategy and FuelCell Energy's distributed power solutions suggests that the next phase of AI infrastructure will be less about centralized hyperscale and more about distributed, regionally optimized capacity.