Argentum AI has closed a $2.5 billion three-party agreement with cloud gaming operator Boosteroid and real estate developer DL Invest Group for a 300-megawatt data center deployment in Europe. The deal structures shared infrastructure between gaming edge distribution and AI inference workloads, a configuration that addresses both parties' power density requirements without separate builds. The facility location remains undisclosed, though Boosteroid's existing concentration in Eastern Europe and DL Invest's portfolio tilt toward Poland and Romania suggest a Central European footprint. Argentum AI takes operational control of the compute layer while Boosteroid secures long-term rack access under what appears to be a capacity-reservation model rather than straight lease.
The 300-megawatt figure puts this facility in the upper quartile of European hyperscale deployments announced in 2025. That power envelope supports roughly 75,000 to 90,000 GPU equivalents depending on cooling architecture and chip generation, though the actual mix will likely blend lower-density gaming servers with higher-density AI training clusters. Boosteroid's business model—streaming AAA titles to consumer devices—requires sub-20-millisecond edge latency, which limits rack placement to metro fiber rings. Argentum AI's inference workloads tolerate slightly higher latency but demand burst capacity for model fine-tuning, creating complementary load profiles that smooth utilization curves. The shared capital structure reduces Argentum's upfront build cost by roughly 30% compared to a greenfield solo deployment, while DL Invest captures long-term occupancy guarantees that de-risk construction financing.
This deal matters because it shows AI workload aggregators now treating gaming infrastructure as interchangeable compute substrate. Boosteroid operates over 4 million active users across 15 countries, mostly in markets where console penetration is low and fiber broadband density is high. That footprint overlaps precisely with the regions where European GPU supply remains constrained and power permitting lags demand by 18 to 24 months. Argentum AI gains immediate access to live production capacity without waiting on grid connection queues, while Boosteroid monetizes off-peak rack space during non-gaming hours—typically 06:00 to 14:00 UTC when European user activity troughs. The revenue share structure, not disclosed in the announcement, likely favors Boosteroid during prime evening windows and Argentum during daytime inference runs. Multi-tenant data centers have existed for years, but purpose-built dual-use facilities optimized for both edge gaming and centralized AI workloads represent a new capital formation model that could reshape how European compute capacity gets financed and allocated over the next 36 months.
Allocators should track three near-term indicators. First, whether Argentum AI announces customer logos for the European inference capacity within 90 days—unnamed capacity suggests speculative build rather than committed demand. Second, if Boosteroid files for expanded rack commitments beyond the initial deal, which would confirm the shared-utilization model performs operationally. Third, whether DL Invest packages similar structures with other AI operators, turning this into a repeatable financing vehicle rather than a one-off arrangement. The company has previously developed logistics and industrial real estate in Poland and the Czech Republic but no prior hyperscale data center exposure, making execution risk non-trivial.
European power costs for 300-megawatt continuous draw currently run €40 to €65 per megawatt-hour depending on grid connection terms and renewable purchase agreements, translating to roughly €105 million to €170 million in annual energy expense before cooling overhead. The capital structure absorbs that through tenant billing, but margin compression in European cloud services over the past 18 months means pricing discipline will determine whether this facility operates profitably or becomes a stranded asset if either anchor tenant underperforms.