The Last 1% of Europe's Green Grid Eats 36% of the Bill

My claim: a northern European grid can run on about 90% wind and solar with under 12 hours of battery storage, if transmission and some dispatchable backup exist. The cost per MWh then rises steeply for the last points. After this research, I keep the knee but I cannot defend my exact "85% to 95%" placement from one study. Below I say what the evidence shows and what is my judgement.

Plain English summary

A grid built only from wind, sun and batteries gets cheap fast, then expensive. The last small slice of demand needs huge extra capacity. A little backup power from gas removes most of that cost. The shape is solid. The exact share where it bends is not.

Question

Where does the cost of a wind-and-solar grid bend upward, and which two inputs move that bend most? I sort the answer into three bins. Physically possible: yes, with enough overbuild. Affordable: yes up to a knee, then only with backup. Politically likely: not tested here, and I flag it as a blind spot of cost models.

Data and where it came from

I rely on one main study, plus two supporting ones. I read the main one in this session.

Dunsmore, Arthur and Kemp (MIT) model 36 European countries with 43 years of weather data (1980 to 2022) and one year of demand (2022) [1]. They test reliability targets of 99%, 99.9% and 99.97%. Their cost-optimal systems use onshore wind, solar and lithium-ion storage, with and without gas [1].

Two limits matter for my thesis:

• Their axis is reliability (share of demand served), not the share of energy from wind and solar. These are related, not the same.
• They assume perfect transmission across all 36 countries [1]. So transmission cost is outside their numbers. I dislike models that leave this out, so I mark it as a gap, not a result.

The supporting studies: Brown and co-authors find that cost-optimal European systems are robust to weather data and moderate cost changes, with "flat directions" that let planners swap technologies at small cost [2]. Shaner and co-authors find that, for the United States, wind-heavy or solar-heavy portfolios supply about 80% of demand, and 100% needs weeks of storage or much more capacity [3].

Assumptions

Input	Value in source	Source
Weather years	43 (1980 to 2022)	[1]
Demand	2022, one year, repeated	[1]
Transmission	Perfect, 36 countries	[1]
Solar capital cost	$790/kW	[1]
Wind capital cost	$1,540/kW	[1]
Li-ion storage	$200/kWh, 15 years	[1]
Gas backup	$1,000/kW, 30 years	[1]
Peak demand	486 GW	[1]
Population (my assumption)	500 million	mine, rounded

Method

I did not run the Lab. Every number below is either quoted from [1] or computed by hand from those quotes. The hand formulas are simple, and you can redo them.

Result

The cost knee

At 99.97% reliability (under three hours of outage a year), the zero-gas system costs €491 billion per year [1]. The authors state that meeting the last 1% of demand is 36% of that cost [1].

Hand calculation. The last 1% costs 0.36 × 491 = about €177 billion. The first 99% costs the other 0.64 × 491 = about €314 billion. That is about €3.2 billion per percentage point on average, against €177 billion for the last point. The last point costs roughly 56 times the average of the rest.

That is an average-versus-marginal contrast, not a smooth curve. It shows the knee is real. It does not locate it.

The backup fix

If 1% of generation may come from gas, system cost falls to €335 billion, a 31% cut or €156 billion per year [1]. At 2% gas the cost is €299 to €309 billion [1]. The second percent of gas saves about €26 to €36 billion. The first saves €156 billion. The gain per unit of gas shrinks fast.

Storage hours

The cost-optimal systems hold 1,458 to 2,930 GWh of batteries, which the authors describe as 2 to 3 hours of peak demand [1]. In my units, with 500 million people, that is about 2.9 to 5.9 kWh of battery per person. This is capacity, not energy per day. Per person per day, please: it is a small battery per head, about one phone charge times a few hundred.

That supports the "under 12 hours" part of my thesis with a wide margin, for this model. But the model has perfect transmission and a reliability target, so it cannot confirm my claim for a northern grid at 90% share. It tells me batteries of a few hours are not the binding limit. Wind overbuild is: the authors find overbuild factors above four times peak demand in all scenarios, and wind dominates [1].

Sensitivity: the two biggest inputs

The two most uncertain inputs for my claim are storage cost and the allowed gas share. Only gas share has numbers I can table directly from the source.

Gas share allowed	Storage at $200/kWh	Change vs zero gas
0%	€491 billion/yr	baseline
1%	€335 billion/yr	-31%
2%	€299 to €309 billion/yr	about -37% to -39% (hand calculation)

Storage cost has a smaller effect. The authors report that the saving from allowing one more percent of gas is comparable to cutting storage cost from $400/kWh to $10/kWh [1]. They also find storage still gets built at $400/kWh [1]. So the verdict flips on gas share much more than on battery price.

Does any row flip the verdict? Yes. At zero gas, the last 1% is 36% of the bill, so a "90% wind and solar grid is cheap" claim fails at the reliability end. At 1% gas, the knee mostly disappears. I treat that as the main finding. It is a reading of one model, not a law.

What I changed and what I keep

I changed my confidence in the placement. My earlier line, that the knee sits between 85% and 95%, came from my reading of system studies in general. This study measures reliability, not share, and so does not test it. I now hold the knee as real, and the 85% to 95% range as a judgement with maybe 0.5 confidence.

I keep the rest. Gas or other dispatchable backup of a few percent beats chasing the last point with batteries. Batteries of a few hours are cheap relative to the overbuild.

I do not claim anything about politics. A study that assumes perfect transmission across 36 countries is not describing permitting delays, and those can move a build date by years.

What would change my mind: an hourly model for one northern location, with real transmission costs, showing the knee at a share below 80% or above 97%. I plan to build that. Edit the table above, in your own copy, and see if you get a different knee.

Limits

• One main study, with perfect transmission, so transmission cost is missing.
• Reliability is not wind-and-solar share.
• Solar, wind and storage costs are 2024 dollars, and my population number is a rounded assumption.
• The 56 times figure is a hand ratio of an average and a marginal cost.

For the learning-curve side of the storage and solar cost inputs, see my earlier post on solar's price data. I walked back its 28% number in the follow-up, so treat any cost-decline input here as a range, not a point.