PV + Storage Is NOT All You Need: A Reality Check on Off-grid Data Centers

This repository replicates and extends Casey Handmer's solar+battery optimization model, adding the real-world costs he omitted (O&M, land, battery replacement, degradation) and comparing against grid-connected alternatives across multiple regions. We find that for high-reliability loads like data centers, the grid wins at high latitudes -- and Handmer's CapEx-only model significantly understates the true cost of off-grid power.

Full PDF report: output/solar_storage_scandinavia_report.pdf

With real-world costs included, grid beats off-grid for data center loads in Denmark:

Left: Total system cost per unit utilization. The grid (green) crosses below off-grid (red) at ~EUR 7M/MW load CapEx -- right in the data center range. Right: Off-grid utilization plateaus at 75-90%, far below the 99.9% data centers require.

Extending Handmer's analysis across all his regions, plus Denmark:

Handmer's original results (US locations + Britain) cluster together. Denmark off-grid (red) tracks above them due to lower solar yield. The Danish grid (green) undercuts off-grid for loads above ~EUR 7M/MW. Note: Handmer's results use CapEx-only costs; our Denmark results include full lifecycle costs.

The fair comparison: size the off-grid system to match the grid's reliability (99.9% uptime), then compare costs. Results for a 200 MW data center (typical 2025-2026 hyperscaler facility, EUR 5M/MW CapEx = EUR 1B total):

Denmark (57N, solar CF 0.13, wind CF 0.30):

North Texas (36N, solar CF 0.19, wind CF 0.40):

Key findings:

• Off-grid at 99.9% is cheaper than grid in both locations -- but requires enormous infrastructure (800MW of generation capacity + 1+ GWh of batteries for a 200MW load)
• Wind is essential -- solar-only systems cannot achieve 99.9% in Denmark at any cost within reasonable bounds
• The jump from 99% to 99.9% is extremely expensive (battery costs dominate)
• Grid connection costs ~EUR 54M once; the rest is 25 years of electricity at spot+tariffs
• Texas grid is cheaper than Denmark grid (lower spot prices, lower tariffs)
• Real-world feasibility questions remain: siting 800MW of generation, permitting, land availability, and whether synthetic wind/solar profiles capture worst-case weather

On August 6, 2025, Casey Handmer (@CJHandmer) published a deep analysis arguing that rapidly falling solar and battery costs make off-grid power systems viable for virtually any load, anywhere -- dismissing grid backup as unnecessary. He shared his data and Mathematica code on August 7.

His key claims:

• Solar at EUR 200k/MW and batteries at EUR 200k/MWh make off-grid viable
• Massive solar overbuild is cheaper than large batteries for cloudy days
• Geography barely matters (factor of 2 cost variation across 6 locations)
• Grid connections are unnecessary ("DEI for turbines")

We test these claims at 57N latitude (Denmark) where solar yield is 35% lower than Texas, and compare against actual grid connection costs and DK1 spot electricity prices.

Handmer's model considers only upfront capital expenditure: solar panels and batteries. Once purchased, the system runs for free forever. This is the single largest omission in his analysis and systematically biases results in favour of off-grid.

Our model adds the costs that any real project must bear:

For a 10 MW solar array with 12 MWh battery over 25 years, these "missing" costs add approximately EUR 4-5M to the off-grid system cost -- roughly doubling the power system cost relative to Handmer's CapEx-only model.

Handmer does not model a grid-connected alternative at all. His framework only optimizes solar array size and battery size for a standalone off-grid system. The implicit assumption is that the grid either does not exist or is too expensive to consider.

We add a grid-connected scenario using:

• Grid connection fee: 2,000,000 DKK/MW (~EUR 268k/MW), one-time
• Electricity: Historical DK1 spot prices (2023, mean ~87 EUR/MWh)
• Grid tariffs: 20 EUR/MWh (60kV DSO) or 16 EUR/MWh (TSO), based on 2026 Energinet rates. Includes system tariff, network tariff, DSO capacity charge, and electricity tax (business rate). PSO was abolished in 2022.
• 100% utilization: The grid is always available

This allows a direct comparison: what does 1 MWh of delivered energy actually cost from off-grid solar+storage vs from the grid?

Handmer's analysis has no grid costs because he models no grid scenario. We researched actual grid tariffs for large industrial loads (10-100+ MW) across all regions in his analysis:

Britain stands out with grid tariffs roughly 10x those of Texas. This actually strengthens Handmer's case for the UK, where the grid is so expensive that off-grid solar may be competitive despite mediocre solar resources.

Handmer tested 6 locations (32-52N latitude range) and found "minimal variation." Our Denmark site at 57N exposes the limits of this claim:

• Annual yield: Denmark 1,100 kWh/kWp vs Texas 1,700 kWh/kWp (35% less)
• Winter crisis: At 57N, December days are 6-7 hours with very low sun angle. Multi-day cloud cover is routine. The darkest weeks produce <5% of summer output.
• Seasonal storage problem: Batteries sized for overnight storage (12-16h) cannot bridge multi-day or multi-week winter deficits. You would need weeks of storage, which is economically absurd with lithium-ion at EUR 200k/MWh.

Handmer optimizes "cost per unit utilization" -- total system cost divided by the fraction of the year the load runs. This metric implicitly treats partial utilization as acceptable. For an electric kettle or water pump, it is. For a data center where each percentage point of downtime on a EUR 50M/MW facility represents ~EUR 500k/year in stranded capital, it is not.

Our analysis tracks utilization explicitly and shows that off-grid systems in Denmark plateau at 75-90% utilization for data center loads -- far below the 99.9%+ required for critical infrastructure.

Handmer uses NREL's Solar Power Data for Integration Studies: measured/simulated 5-minute resolution data from ~6,000 PV plants across Texas (2006 weather year). For his other locations (Arizona, Britain, California, Maine, Washington), he published only the pre-computed optimization results, not the raw solar profiles.

We generate synthetic 5-minute solar profiles using pvlib (open-source solar modelling library) with:

• Ineichen clear-sky model
• Monthly clearness indices calibrated to known annual yields
• AR(1) cloud variability process with beta-distributed cloud factors

Our synthetic Texas profile (CF 0.192, 1,685 kWh/kWp) validates well against Handmer's NREL data (CF 0.191, 1,673 kWh/kWp).

Texas, Arizona, Washington -- These sun-rich regions with low grid tariffs (EUR 9-20/MWh) and high solar yields (1,500-1,900 kWh/kWp) are where Handmer's thesis is strongest. Off-grid solar+storage costs are low, utilization can reach 95%+, and the grid's cost advantage is slim. For loads that can tolerate occasional curtailment, off-grid is viable.

California -- Despite excellent solar resources (~1,800 kWh/kWp), California's extraordinarily high grid tariffs (~75 EUR/MWh in non-generation charges) make off-grid solar economically attractive even for loads requiring high utilization. California is perhaps the strongest case for Handmer's thesis in a developed economy.

Britain -- Counterintuitively, Britain's terrible grid economics (115 EUR/MWh in non-commodity charges) make off-grid solar competitive despite mediocre solar resources (~1,000 kWh/kWp). The grid is so expensive that even inefficient off-grid systems may be cheaper. However, utilization will be low (similar seasonal problems to Denmark), making this viable only for loads that tolerate intermittency.

Denmark / Scandinavia -- With moderate grid tariffs (~16-20 EUR/MWh), reasonable spot prices, and poor winter solar resources, the grid wins for high-CapEx continuous loads above approximately EUR 7M/MW. This includes essentially all data center applications. The crossover point is clearly visible in our optimization results.

Maine / New England -- Despite relatively high grid tariffs (~40 EUR/MWh), the solar resource at northern latitudes (similar seasonal issues to Scandinavia) and high reliability requirements favour grid connection for continuous loads.

Even where grid connection is the economically optimal solution, connection timelines can be a critical blocker. In Denmark and across Europe, grid connection for large loads (50-100+ MW) routinely takes 3-7 years due to:

• Grid capacity constraints: Transmission and distribution networks in many regions are at or near capacity, requiring reinforcement before new large loads can connect
• Permitting and environmental review: New substations and transmission lines face lengthy planning processes
• Queue backlogs: In many markets (notably ERCOT, PJM, and European TSOs), the interconnection queue has grown to years of backlog as data centre and renewable energy projects compete for limited grid capacity
• Equipment lead times: High-voltage transformers and switchgear have 18-36 month delivery times

This means that even if the grid is cheaper in steady-state, a data centre developer who needs power in 12-18 months may have no choice but to deploy solar+storage as a bridge solution -- or as the primary power source if the grid connection never materialises.

Handmer's thesis gains significant practical weight in this context: not because off-grid is cheaper, but because it is faster to deploy. A solar+battery system can be installed in 6-12 months. In a market where time-to-power determines whether a project happens at all, the grid's theoretical cost advantage is irrelevant if it comes with a 5-year wait.

generate_solar_data.py    # Synthetic solar profiles via pvlib (Denmark, Texas)
optimize.py               # Off-grid, grid-connected, and hybrid optimization
generate_report.py        # PDF report with plots
data/                     # Solar profiles, spot prices, optimization results
data/handmer/             # Casey Handmer's original data and results
output/                   # PDF report and plot images

python -m venv .venv && source .venv/bin/activate
pip install numpy pandas scipy matplotlib pvlib fpdf2

python generate_solar_data.py   # Generate 5-min solar profiles
python optimize.py              # Run all optimizations
python generate_report.py       # Generate PDF report

• Solar profiles: pvlib synthetic (Denmark), NREL Solar Integration Study (Texas)
• DK1 spot prices: Energi Data Service API (2023 historical)
• Grid tariffs: Energinet (2026 rates), ERCOT, NESO, EIA state data
• Handmer's data: Google Drive

This analysis is provided for educational and research purposes. Handmer's original data is shared under his terms. Our code is MIT licensed.