Polarise plans 30 MW AI data center in Bavaria to boost European sovereignty
Polarise is planning a new AI-focused data center in Amberg (Unterallgäu, near Augsburg), with an initial 30 MW and a path to 120 MW. The company is targeting mid-2027 for the first phase to go live. The bet is clear: more AI compute under European control, less dependency on non-EU hyperscalers.
The site is the former Wertachtal shortwave transmitter. Polarise aims to acquire the building and land from a consortium including Stadtwerke Bad Vilbel and a local civic foundation. According to the consortium's representative, Klaus Minkel, permits will be filed after the local city council elections.
Why this matters for IT and development teams
European AI workloads need predictable latency, compliance with EU rules, and data locality. A domestic, large-scale facility helps teams meet those constraints without routing sensitive training and inference across borders. According to industry association Bitkom, Germany had ~530 MW of AI data center capacity at the end of last year, much of it run by global hyperscalers.
Modular "AI Pods" to compress time-to-capacity
The build will lean on Polarise's AI Pods-modular HPC clusters that slot into existing data center infrastructure. The company points to its Munich project, which went from concept to commissioning in six months, versus the typical 2-3 years for new builds. Expect a faster ramp to usable GPU capacity, staged by cluster.
Energy design: renewable-first with storage
Polarise is working with WV Energie AG (Bad Vilbel), which already operates on-site solar parks totaling ~70 MWp. Battery storage is planned to buffer demand and improve availability. The roadmap adds a new PV park up to 35 MWp, a 21 MW wind farm, and combined heat and power-together enabling up to ~120 MW of green energy over time.
The stated goal: show that high-density AI compute and the energy transition can reinforce each other when grid, storage, and generation are planned as one system.
Scale and market context
At full buildout, Amberg would sit among Germany's larger AI facilities. For reference, hyperscalers like Google and AWS commonly deploy in 100 MW blocks or more. Polarise currently operates 13 data centers in Germany and abroad and plans to expose AI compute via its own cloud platform-while keeping infrastructure under European control.
Timeline, costs, and what's public
Phase one targets mid-2027 operations at 30 MW. The company hasn't disclosed the investment amount. A source close to the project indicates a spend in the hundreds of millions of euros for base infrastructure, excluding AI chips. Final costs will depend on the mix and count of accelerators customers bring online.
What technical leaders should plan for
- Workload placement: Training vs. inference tiers, data residency, and proximity to sovereign datasets.
- Capacity planning: Start with 30 MW, expect staged scale toward 120 MW; plan contracts that allow flexible GPU ramp-up.
- Network: Multi-100 Gbit/s fabric inside pods and low-latency interconnect between clusters-design your data pipelines accordingly.
- Energy alignment: Understand green energy SLAs, battery-backed availability windows, and how your workloads map to variable supply.
- Procurement: Separate facility/cluster contracts from accelerator SKUs; costs swing wildly based on chip class and utilization profile.
- Compliance: Map EU data rules and AI governance to deployment choices; keep sensitive training data local where required.
Key numbers at a glance
- Initial capacity: 30 MW (mid-2027 target)
- Planned scale: Up to 120 MW
- On-site renewables: ~70 MWp solar existing; +35 MWp PV planned; 21 MW wind planned
- Storage: Battery systems planned to stabilize consumption
- Build model: Modular AI Pods; prior project delivered in ~6 months
- Market context: ~530 MW AI DC capacity in Germany overall (Bitkom), largely run by international providers
- Investment: Hundreds of millions of euros for phase one base infra (excluding AI chips), per a source close to the project
Why it's strategically useful
Sovereign AI isn't just a policy slogan-it dictates where your models live, who operates the stack, and how you meet regulatory audits. A domestic, modular site like Amberg gives teams another option for high-density compute with a renewable-heavy energy profile. If you're planning GPU expansion in the EU, put this on your short list.
Your membership also unlocks:
