Bottom line: Southeast Europe (SEE) is absorbing more hyperscale capital in 2026 than in the prior decade combined, as FLAP-D congestion pushes AI-driven demand toward secondary markets. But the region’s pipeline is a mix of financially committed projects, signed MoUs, and press-release ambition – and the gap between those categories, not tenant demand or land availability, will determine how much of the announced ~3+ GW actually gets energized this decade. Power grid capacity and the speed of substation buildout are the binding constraints; everything else is second-order.
Why capital is rotating to SEE
Primary European hubs are effectively full: grid interconnection queues, land scarcity, and permitting timelines in Frankfurt, London, Amsterdam, Paris, and Dublin have pushed vacancy below 8% and lead times for grid connection well past what hyperscalers are willing to tolerate. SEE offers three things those markets can’t: a materially lower land and labor cost basis, new subsea cable capacity (Black Sea landings at Constanța, Romania, and Mediterranean routes through Greece via BlueMed, GreenMed, and 2Africa), and largely untapped wind and solar potential that can support the hourly-matched renewable PPAs hyperscalers increasingly require for AI workloads.
Why it isn’t a sure thing
Three risks are structural, not incidental:
Power and grid firming. Announcing a campus and energizing it are different problems. Utility-scale sites need dual high-voltage feeds and firm baseload; grid congestion and multi-year substation lead times are the single largest source of schedule risk in every SEE market.
Talent. The region has no dedicated pipeline of liquid-cooling and critical-facility engineers, and specialized academic programs are largely absent.
Execution. Gigawatt-scale projects slip 12–18 months even in mature FLAP-D markets. SEE’s subcontractor pools and supply chains are largely untested at this scale, which should push risk-adjusted timelines further right, not hold them where developers state them.
The EU/non-EU split is the real organizing variable
EU member states (Croatia, Romania, Greece, Bulgaria) benefit from regulatory harmonization – GDPR and the EU AI Act are pushing US tech firms to host European data inside EU borders – and from direct capital, like Romania’s €500 million NRRP allocation to sovereign cloud infrastructure. Non-EU markets (Serbia, North Macedonia) have no equivalent subsidy pool, so governments are acting as anchor tenants and de-risking projects directly, while racing to reform permitting law to compensate. That works for attracting announcements; it does less to solve the harder problem non-EU markets face, which is assembling a bankable, ESG-compliant energy stack that Western hyperscalers will actually sign long-term contracts against.
Exhibit 1: Announced 10 MW pipeline by country (as of August 2026)
| Country | Project | Investment | Capacity | Status |
| Croatia | Pantheon AI (Topusko) | €50B total (€12B phase 1) | 1 GW total / 800 MW IT, Tier IV+ | Announced Apr 2026; construction targeted 2027, operations 2029 |
| Croatia | NEOIX AI Center (Nova Gradiška) | €450M | 50 MW | Preliminary agreement signed Dec 2025 |
| Croatia | Digital Realty – ZAG1 (Zagreb) | Undisclosed | +30 MW white space | Ongoing expansion |
| Romania | ClusterPower / AIC (Mischii + Fauresti, Craiova) | Undisclosed | 800 MW phased through ~2029 | Equity partnership signed late 2025/early 2026 |
| Romania | Market-wide colocation | – | ~93 MW (2026) → ~232 MW (2031)¹ | 19.9% CAGR per one tracker – see note below |
| Greece | PPC Kozani Mega AI Hub (W. Macedonia) | €5.75B regional plan² | 300 MW phase 1 → up to 1 GW | Figures still moving – see note below |
| Greece | Microsoft “GR for Growth” | Undisclosed | 3 facilities | Ongoing (Spata / Koropi) |
| Greece | Data4 Campus (Paiania) | €300M | 90 MW | Broke ground late 2024; phase 1 live late 2026/early 2027 |
| Serbia | State Data Centre (Kragujevac) | €50M (2nd supercomputer) | 14 MW today; +40 MW planned | 2nd supercomputer live Apr/May 2026; e& enterprise MoU (Sep 2025) targets a 40 MW extension block; 3rd supercomputer (Bull SAS / Mistral AI) due 2027 |
¹ One widely cited estimate (Mordor Intelligence) puts Romania’s 2026 colocation base at 93 MW growing to 232 MW by 2031. A separate industry tracker (EUDCA) puts the 2025 base at just 27 MW growing to 66 MW by 2031. The four-fold difference in stated base capacity is a useful reminder that SEE market sizing is still unstandardized – treat any single-source figure as directional, not precise.
² PPC’s own public figures have moved across 2025–2026 disclosures: the Kozani data center alone has been costed anywhere from €2.3B to €8B in different statements and press reports, while €5.75B is PPC’s figure for the combined data-center-plus-energy regional investment plan. Use the €5.75B figure as the regional total, not the data center line item.
What this means, by stakeholder
For hyperscale tenants: the region offers real cost and latency advantages, but power-on dates on anything announced in the last 12 months should be discounted by at least a year against operator guidance – grid interconnection, not construction, is the pacing item across every market in Exhibit 1.
For infrastructure investors: the EU/non-EU split is a proxy for financing risk. EU-market projects can lean on regulatory harmonization and NRRP-style capital; non-EU projects (Serbia, North Macedonia) depend on state anchor-tenancy and bilateral MoUs that can stall at the political level, as North Macedonia’s contested construction-law reform illustrates.
For governments: the non-EU playbook – fast-track permitting plus state-backed anchor demand – is necessary but not sufficient. Without a comparably fast path to bankable, ESG-compliant power, permitting speed alone won’t close deals with hyperscalers who have EU-market alternatives.
Outlook
SEE now has three genuine gigawatt-scale initiatives on the board – Croatia’s Pantheon AI, Romania’s ClusterPower/AIC, and Greece’s PPC Kozani – which is a real inflection from a colocation-only market few years ago. Whether the region converts announcements into energized capacity at anything close to stated timelines depends less on any single deal and more on transmission operators’ ability to deliver certified green baseload on a schedule that, so far, no SEE market has demonstrated at scale.