HPC & AI Factory — Land Acquisition Site Analysis
Avram Iancu 369MWac PV
+ 1,475 MWh BESS
24-Factor Scoring Matrix assessment for hyperscale data center and AI factory co-location potential. Bihor Region, Romania — prepared by S.C. Cartesio SRL, Bucharest. Evaluated against the institutional-grade site selection framework for 100MW–1GW+ compute facilities.
Overall Tier
B+
Conditional — address connectivity gaps
Composite Score
3.28
/ 5.00 weighted
Recommendation
CONDITIONAL
Proceed with specific remediation plan
Primary Use Case
POWER HUB
Behind-the-meter RE + BESS for AI training
Key Strengths (Top 3)
S
Renewable Energy Access: 369MWac on-site solar + 1,475 MWh BESS is a globally rare behind-the-meter clean power configuration. Enables carbon-neutral AI compute from day one with storage-buffered baseload.
A
Grid Capacity (400kV): ATR (authorization to connect) issued April 2025 for 400kV connection. 369MW solar capacity with grid export potential. Shared 400kV substation reduces per-MW connection cost.
A
Land & Expansion: 500 Ha total site with 10 Ha earmarked for station/BESS. Massive expansion room for co-located HPC facility. 30-year lease provides long-term security.
Key Risks (Top 3)
C
Internet Exchange Proximity: Nearest major IXP (RoNIX) is in Bucharest, ~550km away. No local IX in Bihor region. Oradea is nearest metro but lacks DC-grade IX infrastructure. Critical gap for inference workloads.
C
Workforce & Talent Pipeline: Avram Iancu is a small rural municipality. Oradea (~200K pop, 50km) provides regional talent but is not an established tech hub. University of Oradea has limited CS/EE programs. Recruitment premium required.
B
Latency Profile: Estimated 15-20ms RTT to Budapest, 25-30ms to Vienna/Frankfurt. Acceptable for AI training batches but marginal-to-poor for real-time inference SLAs. Site is positioned as "power hub" not "latency hub."
Disqualifier Override Check
Power: PASS — 369MW solar + 400kV grid connection confirmed via ATR. Path to 100MW+ within 36 months is clear.
Fiber: CAUTION — No IXP within 100km, but Oradea (50km) has backbone fiber operators (DIGI/RCS&RDS, Orange, Telekom). Not a hard disqualifier given 93% national fiber coverage, but requires dedicated dark fiber build-out for DC-grade connectivity.
Flood: PASS — Bihor highlands, elevated terrain. Not in 100-year flood zone based on Wody Polskie/Romanian ANAR equivalents.
Seismic: PASS — Western Romania (Bihor) is low-seismic. Vrancea zone (main risk) is 500km+ east. PGA <0.1g.
Zoning: CAUTION — Currently zoned/permitted for solar PV, not for data center. Would require amendment to planning for DC co-location. 12-24 month conversion timeline estimated.
Political: PASS — Romanian government actively pursuing DC sector growth. EU Modernisation Fund support for renewables confirmed.
DC Installed Capacity
~93 MW
59 facilities nationwide. 231 MW projected by 2031 (19.9% CAGR)
DAM Wholesale Price
€97–148/MWh
Q1 2025: €148/MWh. Sept 2025: €97/MWh. Feb 2026: 519 lei/MWh (~€104)
Fiber Coverage
93%
National locality coverage (FTTH/B). DIGI, Orange, Telekom backbone
Hyperscaler Presence
Emerging
Google MoU signed. No AWS/Azure region yet. ClusterPower 200MW in Dolj
RE Mix
50%+
Hydro 6.14GW + wind + solar. Nuclear expansion (Cernavoda life extension)
2030 RE Target
30.7%
7 GW new capacity planned, of which 3.7 GW solar. €3B EU Modernisation Fund
1
Grid Capacity (MW)
A
369MWac solar capacity with ATR (nr 2374/02.04.2025) confirmed by Transelectrica. 400kV connection to Oradea Sud–Nădab line. For DC use, ~100-200MW could be redirected from solar generation during peak hours or drawn from grid. Not S-tier because allocation is for generation, not load — would need separate consumption ATR for HPC co-location.
2
HV Grid Proximity
A
400kV line runs through/adjacent to the site (Oradea Sud–Nădab corridor). Dedicated 400kV substation planned on 10 Ha plot near the line. Shared cost with co-developer reduces CAPEX. Substation construction pending but funded. Strong A-tier — on-site 400kV is rare for greenfield sites.
3
Grid Redundancy (N+1)
B
Single 400kV connection planned. No evidence of dual independent feeds or separate substation backup. Romania's western grid (Transelectrica) has moderate redundancy through Hungarian interconnection (5km to border), but on-site N+1 would require additional investment. BESS provides partial backup (1,475 MWh = ~4 hours at 369MW).
4
Power Cost (€/MWh)
A
Behind-the-meter solar PV levelized at ~€30-40/MWh (zero grid charges). Business plan base case assumes €50/MWh merchant sales. For co-located DC, effective power cost could be €35-50/MWh via internal PPA — significantly below Romania's wholesale DAM of €97-148/MWh and well below the €60/MWh S-tier benchmark. BESS enables peak-shaving arbitrage adding €10-15M/yr revenue.
5
Renewable Energy Access
S
369MWac on-site solar PV (359 MWp) + 1,475 MWh BESS is an exceptional behind-the-meter RE configuration. Production: 1,234 kWh/kWp/yr (PvGIS validated). Romania's CfD scheme (€78/MWh guaranteed floor) provides downside protection. Bihor receives strong irradiation for CEE (~1,200-1,300 kWh/m²/yr). Best-in-class globally for RE integration with DC.
6
Power Scalability Path
B
Current 369MW is solar generation; scalability for additional consumption load depends on Transelectrica grid reinforcement. 400kV backbone provides headroom. Co-developer building parallel solar project enables potential aggregation. Path to 500MW+ load would require new grid studies and likely 3-5 year TSO planning cycle. Romania's grid modernization is underway but uneven in rural areas.
Power Category Average(4 + 4 + 3 + 4 + 5 + 3) / 6 = 3.83
Weighted Contribution (× 0.30)1.15
7
Fiber Backbone Access
B
Oradea (50km) serves as nearest fiber hub with DIGI/RCS&RDS, Orange, and Telekom Romania backbone presence. Romania has 93% national fiber coverage, but Avram Iancu is rural — last-mile fiber to site would require 20-50km extension. Multiple backbone operators reachable but not within 5km. Route diversity limited — likely single corridor from Oradea.
8
Internet Exchange Proximity
C
RoNIX (Romania's primary IX) is in Bucharest, ~550km away. No local IXP in Bihor. Nearest significant peering points are in Budapest (BIX, ~180km via Hungary) or Vienna (~500km). For DC-grade connectivity, this is a significant gap. Would need private peering arrangements or satellite IX extension. Not viable for latency-sensitive inference.
9
Dark Fiber Availability
C
No confirmed dark fiber inventory near Avram Iancu. DIGI network reaches Oradea but lit fiber, not dark. Would require procurement of new dark fiber build from Oradea (50km) or potentially cross-border to Hungary (5km to border). Estimated cost: €1-3M for dedicated dark fiber link. Lead time: 12-18 months.
10
Subsea/Cross-Border Links
B
5km from Hungarian border provides relatively fast path to Budapest (BIX) and onward to Vienna/Frankfurt via established terrestrial routes. Romania's Black Sea submarine cables (Constanța) are 600km+ away. Cross-border fiber to Hungary is the most viable connectivity strategy — Budapest is a Tier-2 European hub with strong peering.
11
Latency Profile
C
Estimated RTT: Budapest ~15-20ms, Vienna ~25-30ms, Frankfurt ~30-40ms. No local IX means all traffic routes through distant peering points. Acceptable for AI training workloads (latency-tolerant batch processing), but disqualifying for real-time inference requiring <5ms to IX. Site is definitively a "training farm" not an "inference hub."
Connectivity Category Average(3 + 2 + 2 + 3 + 2) / 5 = 2.40
Weighted Contribution (× 0.15)0.36
12
Land Area & Expansion
S
500 Ha (5,000,000 sqm) is massively oversized for AI factory co-location at 400-800 sqm/MW. Even dedicating 10% of site to DC (50 Ha) would support 600-1,250 MW of liquid-cooled compute. 30-year lease provides institutional-grade tenure. Contiguous parcels with additional 10 Ha for substation/BESS. Unmatched expansion optionality in Romania.
13
Zoning & Permitting
B
Currently permitted for solar PV generation (ATR issued, building permit Q1/Q2 2025, connection contract Q2 2025). Data center use would require zoning amendment / PUZ (Plan Urbanistic Zonal) modification. Romania's permitting for DC is not yet standardized — no "DC in zoning plan" precedent like Poland's SEZs. Estimated 12-24 months for DC zoning conversion. Municipal support likely given economic impact.
14
Topography & Geotech
A
Aerial imagery shows flat to gently rolling terrain, consistent with agricultural/former industrial land. Suitable for ground-mounted PV (demonstrated) and DC pad construction. No evidence of contamination (greenfield). Standard foundations expected. Bihor region has stable alluvial/sedimentary soils with adequate bearing capacity (>150 kPa typical for region).
15
Transport & Logistics
B
50km from Oradea (E60/DN76). 50km from Arad. Oradea has an international airport (OMR) with limited routes. Budapest Liszt Ferenc (BUD) ~200km provides major international hub access. Road infrastructure is improving (Romania motorway program) but current routes are primarily 2-lane national roads. Heavy transformer delivery (>200t) feasible but would require route engineering.
Land Category Average(5 + 3 + 4 + 3) / 4 = 3.75
Weighted Contribution (× 0.15)0.56
16
Water Supply
B
Bihor region has adequate water resources (Crișul Repede river basin). Municipal supply from Oradea could be extended, but Avram Iancu as a small municipality likely has limited industrial water capacity. Borehole/groundwater is feasible given alluvial geology. Would require dedicated water infrastructure build for DC-scale cooling (500+ m³/day). Investment: €0.5-2M.
17
Water Quality
B
Regional water quality adequate for agricultural use. Industrial/cooling use would likely require treatment (typical for rural Romania). TDS data not available — would need site-specific water quality analysis. Bihor groundwater generally moderate hardness. Treatment capex manageable (€100-300K for on-site RO system).
18
Climate & Free Cooling
A
Bihor region: continental climate, annual average ~10-11°C. Approximately 5,500-6,000 hours/yr below 18°C wet-bulb. Cold winters (-5 to -15°C) provide excellent free cooling for 5-6 months. Summers can reach 35°C+ with heat waves, requiring mechanical cooling. Better than Southern/Eastern Romania but not as favorable as Northern Poland or Nordics. IPCC RCP 4.5/8.5 projections suggest +1.5-2°C by 2050, reducing free cooling hours ~10-15%.
19
Achievable PUE
A
With liquid cooling + free cooling design: annual PUE of 1.15-1.25 achievable. During winter months (Oct-Mar): PUE <1.10 possible. Summer months may push to 1.30-1.35. Annual average well below Uptime Institute global benchmark of 1.55 and competitive with hyperscale best practice. Not S-tier because summer heat limits prevent year-round optimal PUE.
Water Category Average(3 + 3 + 4 + 4) / 4 = 3.50
Weighted Contribution (× 0.10)0.35
20
Seismic Risk
S
Western Romania (Bihor) is in seismic Zone 0-1. The Vrancea deep seismic zone (Romania's primary risk) is 500+ km to the east/southeast. PGA for 475-year return period: <0.08g. Negligible seismic design premium. No active faults within 100km. Romania's seismic risk is concentrated in the Carpathian bend — Bihor is on the stable Pannonian platform.
21
Flood Risk
A
Site is on elevated terrain (aerial imagery confirms). Not located in mapped 100-year or 500-year flood zones based on Romanian ANAR (National Administration for Water) data for Crișul Repede basin. Some flash flood risk in Bihor valleys during spring snowmelt, but site selection on higher ground mitigates. Insurance implications minimal.
22
Natural Disaster Composite
A
Low risk across categories. No mining subsidence (Bihor is agricultural, not mining region). No wildfire proximity (temperate zone). Moderate wind loading (continental, not coastal). No karst/sinkhole geology. Lightning density moderate (~2-4 flashes/km²/yr, typical for continental CEE). Overall insurance profile favorable.
Risk Category Average(5 + 4 + 4) / 3 = 4.33
Weighted Contribution (× 0.10)0.43
23
Workforce & Talent
C
Avram Iancu is a small rural municipality. Oradea (~200K metro, 50km) is the nearest significant city with University of Oradea (limited CS/EE programs). No existing DC/IT employer ecosystem in Bihor for talent poaching. Would need to establish remote operations center in Oradea or offer relocation packages. Romania has strong IT talent nationally (Bucharest, Cluj, Timișoara, Iași) but not in Bihor.
24
Tax Incentives & SEZ
B
Romania CIT: 16% (low by EU standards). Dividend tax: 8% with free capital export. EU Modernisation Fund: €3B approved for renewable support (CfD scheme at €78/MWh floor for solar). No specific DC-focused SEZ program (unlike Poland's PIZ/Investment Zones). R&D super-deduction exists but less generous than CEE peers. Potential for bilateral negotiation with local/regional government for incentive packages.
25
Political & Regulatory Support
B
Romanian government signed Google MoU for DC/cloud. EU AI Continent Action Plan targets tripling DC capacity. National digitalization strategy supports sector. However, no specific fast-track permitting for DC. Regulatory environment still maturing — PPA laws need streamlining, grid permitting complex. Municipal engagement at Avram Iancu level is unknown for DC use case (currently engaged for solar). Score reflects national momentum offset by local uncertainty.
26
Market Proximity & Demand
C
Bihor is ~550km from Bucharest (Romania's DC market center, 58% of capacity). 50km from Oradea (minor market). Nearest significant DC demand: Budapest (~200km, emerging CEE hub). No cloud region proximity (no AWS/Azure/GCP in Romania). Site is not positioned for enterprise proximity — strictly a wholesale/hyperscale power play. Market demand must be created, not captured.
Economic Category Average(2 + 3 + 3 + 2) / 4 = 2.50
Weighted Contribution (× 0.20)0.50
Power (Factors 1-6) — avg 3.83 × 0.301.15
Connectivity (Factors 7-11) — avg 2.40 × 0.150.36
Land (Factors 12-15) — avg 3.75 × 0.150.56
Water (Factors 16-19) — avg 3.50 × 0.100.35
Risk (Factors 20-22) — avg 4.33 × 0.100.43
Economic (Factors 23-26) — avg 2.50 × 0.200.50
COMPOSITE SCORE3.35 / 5.00 → B+ (Conditional)
The Avram Iancu site presents a unique pricing challenge: it is not raw land but a fully permitted, grid-connected, 369MW solar generation facility with 1,475 MWh BESS. The asset's value to an AI factory developer lies primarily in: (a) behind-the-meter clean power at ~€35-50/MWh vs grid rates of €97-148/MWh, (b) the 400kV grid connection (ATR secured), and (c) the massive land bank for DC co-location. Pricing reflects the dual-asset nature (operating RE plant + DC development site).
| Method |
$/MW Estimate |
Rationale |
| Raw Land |
$2,000–5,000 |
Bihor agricultural land ~€3-8/sqm. At 500 sqm/MW (liquid-cooled AI factory density), raw land cost is $1,500-4,000/MW. Romania's rural land is among CEE's cheapest. Leased (not owned) — land rental is €1M/yr for 500 Ha, implying ~€2/sqm/yr lease rate. |
| Powered Land |
$40,000–80,000 |
400kV ATR secured = 4-5× premium over raw land. Behind-the-meter 369MW solar + 1,475 MWh BESS = additional 3-5× premium for clean power at below-market rates. A 400kV connection alone in Romania costs €10-25M and takes 3-5 years. BESS adds ~€150M replacement value. Combined power premium: 8-16× raw land. |
| Replacement Cost |
$50,000–100,000 |
Total project CAPEX: ~€410M (€150M equity + €260M debt). ATR procurement: 2+ years. Environmental permits: completed. Building permit: in progress. Replacing this from scratch: $260K-$1M site costs + €15-25M grid connection + 3-5 year delay. Per MW (369MW base): replacement cost of the power asset alone far exceeds $100K/MW. |
| Time-to-Revenue |
$25,000–60,000 |
Revenue per MW: ~€8-12M/MW/yr (AI training colocation at Romanian power costs). Years saved vs greenfield: 3-4 years. Probability weight: 15% (permits need conversion). Land attribution: 3%. Calculation: $10M × 3.5yr × 0.15 × 0.03 = ~$15,750/MW base, scaled up for BESS value and grid certainty. |
| Comparable Transactions |
$30,000–70,000 |
ClusterPower (Dolj, Romania): 200MW campus, estimated land+power acquisition at €10-20M for 200MW site = $50-100K/MW. CEE DC land (Poland, C&W DC Land Index): powered land at $15-30K/MW for industrial sites. Romania trades at 40-60% discount to Poland. Adjusted: $20-50K/MW. Premium for integrated RE asset: +$10-20K/MW. |
| Fair Value |
$40,000–75,000/MW |
Powered land method dominates. The 400kV ATR, 369MW solar, and 1,475 MWh BESS create a unique "powered campus" with 3-5 year time compression vs any comparable Romanian greenfield. However, a 20-30% discount applies for: (a) connectivity gaps requiring $2-5M fiber investment, (b) zoning conversion risk, and (c) lack of market proximity. At $40-75K/MW × 100MW DC co-location target = $4.0M–7.5M total land/power premium — a rounding error against $1-3B total DC capex. |
The Thesis: Behind-the-Meter AI Training at Romanian Solar Costs
The Avram Iancu site's strongest play is not as a standalone DC site but as a co-located AI training campus powered by captive renewable energy. The economics are compelling:
Power arbitrage: Behind-the-meter solar at €35-50/MWh vs Romania DAM at €97-148/MWh creates a 50-75% power cost advantage. For a 100MW AI training facility consuming ~876 GWh/yr, this equates to €40-80M annual savings versus grid-connected alternatives.
BESS as UPS: 1,475 MWh storage provides ~4 hours of backup at 369MW — functionally equivalent to Tier III+ UPS at zero marginal cost. BESS also enables solar intermittency smoothing, making the "solar-powered DC" actually feasible for 24/7 training workloads when combined with grid backup.
ESG positioning: Carbon-neutral AI training from day one. With Romania's RE mix at 50%+ and on-site solar, this site could achieve PPA-backed carbon neutrality that hyperscalers (Google, Meta, Microsoft) increasingly require of their compute infrastructure.
Revenue stacking: The Cartesio business plan shows base case IRR of 12.65% over 30 years from energy sales alone. Adding DC co-location revenue ($5-10M/MW/yr for AI training) on even 50-100MW of the site would transform the project into a €100M+ annual revenue platform with 25-40% blended IRR.
Critical constraint: This strategy works exclusively for AI training (latency-tolerant). Real-time inference is not viable given the connectivity profile. The investor thesis must be "cheapest clean compute in the EU" not "lowest latency to enterprise."
Total CAPEX
€410M
€150M equity (2%, 30yr) + €260M bank (EURIBOR 15M + 3%, 15yr)
Annual OPEX
€9.0M
O&M €3.3M, insurance €1M, land €1M, storage €2M, other €1.7M
Base Case IRR
12.65%
At €50/MWh over 30-year project life. Best case (€70): 14.43%
Steady-State Revenue
€50-60M/yr
PV energy + storage arbitrage (DAM peak shaving). Year 1 EBITDA: €49M
DSCR
1.24–3.35×
Strong debt service coverage. Year 1 peak (3.35×), stabilizes ~1.24× by Year 14
Production
1,234 kWh/kWp
PvGIS validated. System losses 14%, total losses 21.73%
BESS Lifetime
20 years
1.5 cycles/day × 360 days = 540/yr → ~11,000 cycles total. ABB/Hitachi
30-Year FCF
€700M–1.1B
Cumulative undiscounted. Base case: ~€800M. Best case: ~€1.1B
Strategic Recommendation
CONDITIONAL PROCEED — The Avram Iancu 369MW PV + 1,475 MWh BESS complex scores B+ (3.35/5.00) as an HPC/AI Factory site. This is not a conventional data center site and should not be evaluated as one. Its value proposition is as a captive clean energy campus for AI training workloads where power cost (€35-50/MWh behind-the-meter vs €97-148/MWh grid) is the primary differentiator.
Proceed IF:
1. The intended workload is AI training (not inference). Latency profile disqualifies real-time inference use cases.
2. Investor is willing to fund €2-5M fiber connectivity build from Oradea (50km) or cross-border to Hungary (5km) for backbone-grade dark fiber.
3. Zoning conversion for DC co-location can be secured within 18-24 months (municipal engagement required).
4. The deal structure captures both the RE asset economics (12.65% IRR standalone) and the DC co-location premium — a "solar campus with compute tenant" model, not a pure DC acquisition.
Do NOT proceed if: the thesis requires enterprise proximity, inference-grade latency, or an established local talent pool. For those use cases, redirect to Bucharest (market hub) or Cluj-Napoca/Timișoara (talent hubs).
Comparison to ALTi Stargard (Poland): Stargard scores A+ (4.15) vs Avram Iancu B+ (3.35). The 0.80-point gap is driven primarily by connectivity (Stargard: 3.60 vs Avram Iancu: 2.40) and economic factors (Stargard: 4.00 vs Avram Iancu: 2.50). Stargard's pre-cleared DC zoning, established ENEA relationship, and Polish Investment Zone incentives are material advantages. However, Avram Iancu's on-site 369MW solar + 1,475 MWh BESS is a power asset that Stargard cannot match — making it a compelling portfolio diversification play for a multi-site AI infrastructure strategy spanning CEE.
| # |
Factor |
Tier |
Score |
Category |
| 1 | Grid Capacity | A | 4 | Power |
| 2 | HV Grid Proximity | A | 4 | Power |
| 3 | Grid Redundancy | B | 3 | Power |
| 4 | Power Cost | A | 4 | Power |
| 5 | Renewable Energy | S | 5 | Power |
| 6 | Power Scalability | B | 3 | Power |
| Power Average | | 3.83 | × 0.30 = 1.15 |
| 7 | Fiber Backbone | B | 3 | Connectivity |
| 8 | IX Proximity | C | 2 | Connectivity |
| 9 | Dark Fiber | C | 2 | Connectivity |
| 10 | Cross-Border Links | B | 3 | Connectivity |
| 11 | Latency Profile | C | 2 | Connectivity |
| Connectivity Average | | 2.40 | × 0.15 = 0.36 |
| 12 | Land & Expansion | S | 5 | Land |
| 13 | Zoning & Permitting | B | 3 | Land |
| 14 | Topography & Geotech | A | 4 | Land |
| 15 | Transport & Logistics | B | 3 | Land |
| Land Average | | 3.75 | × 0.15 = 0.56 |
| 16 | Water Supply | B | 3 | Water |
| 17 | Water Quality | B | 3 | Water |
| 18 | Climate & Free Cooling | A | 4 | Water |
| 19 | Achievable PUE | A | 4 | Water |
| Water Average | | 3.50 | × 0.10 = 0.35 |
| 20 | Seismic Risk | S | 5 | Risk |
| 21 | Flood Risk | A | 4 | Risk |
| 22 | Natural Disaster | A | 4 | Risk |
| Risk Average | | 4.33 | × 0.10 = 0.43 |
| 23 | Workforce & Talent | C | 2 | Economic |
| 24 | Tax Incentives & SEZ | B | 3 | Economic |
| 25 | Political Support | B | 3 | Economic |
| 26 | Market Proximity | C | 2 | Economic |
| Economic Average | | 2.50 | × 0.20 = 0.50 |
| COMPOSITE TOTAL | | 3.35 | B+ — CONDITIONAL |
| Site |
Country |
Tier |
Score |
MW Pipeline |
Role |
| Stargard Industrial Park |
Poland |
A+ |
4.15 |
480 MW (→ 1GW+) |
Primary power hub. Grid-connected DC campus. Training + colocation. |
| Avram Iancu PV+BESS |
Romania |
B+ |
3.35 |
369 MW solar + BESS |
Clean power campus. Behind-the-meter AI training. RE portfolio diversification. |
| Bucharest Metro (TBD) |
Romania |
— |
— |
10-50 MW |
Inference hub. Low-latency, enterprise proximity. Requires IX colocation. |
Portfolio-Level Insight
A CEE AI infrastructure portfolio benefits from geographic and functional diversification: Stargard (Poland) as the grid-connected power hub with DC-grade connectivity; Avram Iancu (Romania) as the captive RE campus for carbon-neutral training; and a future Bucharest edge site for inference and enterprise access. This three-node architecture hedges against single-grid dependency, single-country regulatory risk, and workload-type concentration.
The Avram Iancu asset is worth pursuing not as a standalone DC investment but as an integrated RE+compute play within a multi-site CEE strategy. The $4-7.5M land/power premium is de minimis against the power cost savings of €40-80M/yr on a 100MW training deployment.