Electricity is the only major commodity that must be produced at the exact instant it is consumed. There is no warehouse. Supply and demand have to match continuously, every second, or the frequency drifts and equipment starts tripping offline.
For a century that was manageable, because generation was dispatchable — you told a coal or gas plant what to produce and it produced it. Solar and wind are not dispatchable. They produce what the weather allows. Building them is now cheap and fast; the difficulty has moved entirely to the question of what the rest of the system does around them.
That question is what “intermittency” means, and it is worth being precise: it is not one problem. It is four problems with different timescales, different costs and different solutions.
Four problems, not one
| Timescale | The problem | What it looks like | What solves it |
|---|---|---|---|
| Seconds | Inertia and frequency | Spinning turbines physically resist frequency change. Inverter-connected solar and wind do not, so the grid becomes twitchier as thermal plant retires | Synthetic inertia from inverters, synchronous condensers, fast frequency response from batteries |
| Minutes to hours | Ramping | Solar falls away over the evening while demand climbs. The rest of the fleet has to make up the gap at a rate it was never designed for | Batteries, flexible gas, demand response, interconnection |
| Days | Surplus and shortfall | A sunny, windy, mild weekend produces more than anyone needs. A still, overcast week produces almost nothing | Longer-duration storage, pumped hydro, cross-border trade, curtailment |
| Seasons | Adequacy | Northern winters have peak demand and minimum solar at the same time. No amount of four-hour storage bridges December | Firm dispatchable capacity, hydro reservoirs, hydrogen or long-duration chemistries, capacity markets |
Conflating these is the most common mistake in the debate. A battery that solves the evening ramp brilliantly contributes almost nothing to winter adequacy. A capacity market that solves adequacy does nothing for frequency. Each timescale needs its own instrument.
Electricity is the one thing you cannot keep in a warehouse — it has to be made at the exact second it is used. That was manageable while we could tell power stations what to produce. Solar and wind do not take instructions; they produce when the weather allows. The problem is not that they are unreliable. It is that they are unschedulable, and those are different things.
The duck, and what it became
The single most useful picture in power systems is the net load curve — total demand minus whatever wind and solar happen to produce. It is what the dispatchable fleet actually has to serve.
Total demand versus net load, on a spring day
Illustrative shape for a solar-heavy system. The gap between the lines is renewable output
Shape is illustrative, not a specific system's data. The belly deepens every year that solar is added; the ramp steepens with it. In California, solar now supplies close to half of demand between 08:00 and 16:00.
Two things happen as the belly deepens.
Midday power becomes worthless, and then worse than worthless. If more is being produced than consumed and nothing can absorb it, someone must be paid to take it — which is what a negative price is. Generators with subsidies tied to output will keep running through negative prices rather than lose the subsidy, which pushes the price lower still.
And the evening ramp gets brutal. The fleet that has to close that gap runs for a few hours a day at increasing cost per hour, because a plant that starts and stops daily is expensive to run and expensive to maintain.
Draw a day of electricity demand, then subtract whatever the sun and wind happened to provide. What is left is the shape everything else has to fill. As more solar is added, the middle of the day sags towards zero and the evening turns into a cliff. Around lunchtime nobody needs your power; by seven in the evening everybody does, all at once.
What it looked like in 2026
Europe gave a clean demonstration. Second-quarter solar output hit 129 TWh, about 20% above any previous second quarter. The consequences were immediate:
- Spain recorded 596 hours of negative prices in the first half of 2026 — the most in Europe. Portugal had 462. France had 370.
- By late April, exchanges had to lower the minimum price floor from −€500 to −€600 per MWh, because the existing floor was being hit.
- In the same month as those negative prices, German afternoon prices exceeded €600 per MWh on days when the sun went down and demand did not.
Hours of negative day-ahead prices, first half of 2026
The Iberian peninsula has abundant solar and comparatively thin interconnection to the rest of Europe. Denmark and South Australia hit negative net load more often than Germany or Brazil for the same structural reason — small, high-renewable systems with limited ability to export.
California shows the other consequence: curtailment. In 2024, CAISO curtailed 3.4 million MWh — 29% more than the previous year — and 93% of it was solar. That is clean electricity that was generated and thrown away, and it happens mostly in spring, when output is high and cooling and heating demand are both low.
Curtailment and negative prices are not signs of failure. They are signs of a system with more zero-marginal-cost supply than it can currently use — which is the intended destination. The problem is that they wreck the business case for the next solar farm, because the hours it generates are exactly the hours nobody wants power. This is value deflation or capture rate decline, and it is what limits renewable build long before engineering does.
When there is more electricity than anyone wants, the price goes below zero — you have to pay somebody to take it. Spain had 596 such hours in six months. On some of those same days, once the sun set, German prices went above €600. That is not a broken market. It is a market saying very loudly that the valuable thing is no longer electricity — it is electricity at the right moment.
What the ramp actually costs
The evening ramp has to be served by something, and in most European systems that something is a gas turbine. Whether it runs is decided by the clean spark spread — the power price, less the cost of the gas burned, less the cost of the carbon emitted.
Clean spark = Power price − (Gas price × heat rate) − (Carbon price × emissions intensity)
For a modern combined-cycle plant at 55% efficiency, the heat rate is 1.82 megawatt-hours of gas per megawatt-hour of electricity and the emissions intensity is about 0.367 tonnes of CO₂ per megawatt-hour. At August 2026 levels — TTF at €66/MWh and EU allowances at €75 a tonne — the plant needs:
| Power price | Clean spark spread | What the plant does |
|---|---|---|
| Negative, midday surplus | −€150 and worse | Off. Batteries and exports absorb the surplus |
| €90/MWh, ordinary baseload | −€57.6 | Off. Running would lose money on fuel alone |
| €147.6/MWh | €0 | At the margin — the switch-on price |
| €600/MWh, German June evening | +€452.4 | Runs, and earns a large share of its annual margin in a few hours |
That table is the whole economics of a modern peaking plant. It is switched off for the overwhelming majority of the year and makes its money in a few dozen evenings. It is not really a power station any more — it is a strip of call options on the evening ramp, and it should be valued and hedged that way rather than as a baseload asset.
A battery competes for exactly the same scarcity, without a fuel bill. Charge at −€10/MWh in the midday surplus, discharge into the €600 evening at 85% round-trip efficiency, and the economics are:
600 − (−10 ÷ 0.85) = €611.8 per MWh discharged — against the gas
plant's €452.4, on the same evening, with no gas and no carbon to buy.
This is why battery build has outrun every forecast. It is also why the business is self-limiting: every battery added flattens the very spread it lives on. A fleet large enough to fill the ramp is a fleet that has competed away its own revenue — which is why systems that want a lot of storage end up paying for availability through capacity mechanisms rather than relying on arbitrage.
Two things follow for anyone modelling this. First, the carbon price sits inside the switch-on price, so a change in EU allowance policy moves the dispatch order without a single physical change — the arithmetic is set out in Two Carbon Markets. Second, the gas price in that calculation is set half a world away by the LNG market described in The Atlantic–Pacific LNG Arbitrage. A European battery’s revenue in December depends on a shipping lane in the Gulf.
The gas plant that covers the evening is not really a power station any more. It is an insurance policy that sits switched off almost all the time and earns its entire year in a handful of expensive evenings. That is a hard thing to finance — and it is exactly the money a battery is now competing for, without burning anything at all.
The mitigation toolkit
There are six ways to deal with a mismatch between when power is produced and when it is wanted. Storage is only one.
1 · Move it in time
Storage. Charge in the belly, discharge on the ramp. Batteries for hours, pumped hydro for days, other chemistries for longer.
2 · Move it in space
Interconnection. The wind is always blowing somewhere. Transmission is often the cheapest flexibility available — and always the slowest to permit.
3 · Move the demand
Demand response. Only about 100 GW is actively used globally, against 600 GW of residential air conditioning and 160 GW of aluminium smelting that could in principle respond.
4 · Keep firm capacity
Flexible gas, hydro reservoirs, nuclear. Expensive to hold for a few hundred hours a year, which is why capacity markets exist to pay for availability rather than energy.
5 · Throw it away
Curtailment. Genuinely the right answer sometimes — storing every surplus megawatt-hour costs more than losing the few percent that arrive at the worst possible moment.
6 · Change the product
Convert surplus power into hydrogen, heat or desalinated water — anything storable. Round-trip efficiency is poor, so this only works when the input power is nearly free.
There are only six ways to fix a mismatch between when power is made and when it is wanted: move it in time (storage), move it in space (cables), move the demand, keep some old-fashioned plant on standby, throw the surplus away, or convert it into something storable. Storage gets all the attention, but the cheapest answer is very often just a bigger cable.
What batteries actually do
Grid-scale lithium-ion has scaled faster than almost anyone forecast. A record 63 GW was added globally in 2024, taking installed utility-scale capacity to 124 GW, while costs fell about 40% in a single year to roughly USD 150 per kWh. China alone added 42 GW in 2024 and had 145 GW cumulative by the end of 2025.
Penetration is already meaningful in the systems that needed it most: batteries equate to nearly 25% of peak load in California and around 15% in South Australia and Great Britain.
What they are good at is precise and limited:
- Sub-second response. Nothing else on a grid responds faster, which makes batteries the natural provider of frequency regulation.
- The daily arbitrage. Buy in the belly, sell on the ramp. This is the core revenue stream and it depends entirely on the spread between the two.
- Deferring network investment. A battery at a constrained substation can postpone a transmission upgrade for years.
What they are not good at is duration. A four-hour battery is a four-hour battery. It shifts energy within a day; it does nothing for a week of low wind and nothing at all for a season. Durations are lengthening — Australia’s average is expected to rise from 1.5 hours in 2024 to 2.5 hours by 2027, and California concentrates on four-hour systems because its resource-adequacy rules pay for them — but the technology is fundamentally a diurnal tool.
There is also a self-limiting quality to battery economics that is not widely appreciated. Batteries earn on the intraday spread. Adding batteries flattens the intraday spread. A battery fleet large enough to solve the duck curve is a battery fleet that has destroyed its own arbitrage revenue — which is why markets that want a lot of storage end up paying for capacity or availability rather than relying on energy arbitrage alone.
A battery is superb at one job — buying cheap power at lunchtime and selling it dearly at dinner time — and useless at another: getting a country through a still, cloudy week in January. A four-hour battery is a four-hour battery. There is also a catch: every battery you add narrows the very gap it feeds on, so a fleet big enough to fix the problem has competed away its own income.
Pumped hydro and the longer end
Pumped storage hydro is old technology and still by far the largest store of electricity on earth — roughly 9,000 GWh of global capacity against about 120 GW of batteries. Water is pumped uphill when power is cheap and released through turbines when it is dear.
| Technology | Typical duration | Round-trip efficiency | Asset life | Best used for |
|---|---|---|---|---|
| Lithium-ion battery | 2–8 hours | 80–95% | 15–20 years | Frequency response, daily arbitrage, network deferral |
| Pumped hydro | 12–30 hours | 70–82% | 60–100+ years | Multi-day shifting, inertia, bulk capacity |
| Compressed air | 12–24 hours | 42–75% | 40–60 years | Bulk storage where geology allows |
| Flow batteries | 4–12 hours | 65–80% | 20–25 years | Longer daily cycles without degradation |
| Iron-air and similar | 100+ hours | 40–55% | Emerging | Multi-day outage cover, where efficiency matters less than cost |
| Hydrogen | Seasonal | 25–40% | 25–40 years (cavern 80+) | Seasonal balancing — only viable with near-free surplus power |
Indicative industry ranges. Efficiency and duration trade off against each other almost perfectly: the longer you want to store energy, the more of it you accept losing.
Pumped hydro’s advantages are duration, an asset life measured in generations, and real spinning inertia. Its disadvantages are that it needs specific topography, takes the better part of a decade to build, and attracts every planning objection available. Which is exactly why it is chronically under-built relative to what systems need.
Policy is starting to reach past the four-hour battery. Italy’s MACSE auction is contracting 10 GWh of storage by 2028, including 1.3 GWh with durations of eight hours or more. Great Britain’s long-duration scheme targets 2.7–7.7 GWh by 2035. These are small numbers deliberately aimed at a gap the market will not fill on its own.
Pumping water uphill when power is cheap and letting it run back down through turbines when power is dear is a hundred-year-old idea, and it is still by far the largest store of electricity on the planet — roughly seventy times everything held in the world's grid batteries. It lasts for generations. It also needs a mountain, a decade, and a very patient planning department.
India: a very large pipeline and a very small fleet
India is the clearest live case study, and the numbers are striking.
India's storage pipeline versus what is actually running
Gigawatts, as at mid-2026
Of the 10.3 GW running, 7.4 GW is pumped hydro (36 GWh) and 2.9 GW is batteries (8.3 GWh). Source: IEEFA, July 2026.
The economics have moved fast. Standalone battery projects supported by viability gap funding cleared at INR 2.10–2.12 lakh per MW per month in June 2026 — roughly USD 2,200 per MW per month. Renewable-plus-storage packages have been contracted at INR 5.51 per kWh on 25-year terms. Viability gap funding runs up to INR 18 lakh per MWh for standalone projects.
The Central Electricity Authority’s target for FY2030 is 60.63 GW of storage: 41.65 GW of batteries and 18.98 GW of pumped hydro. Getting from 10.3 GW to 60 GW in four years requires the commissioning rate, not the tendering rate, to change — and the constraints are land, grid connection and offtaker credit rather than technology or cost. The 16.4 GW Madhya Pradesh pumped-hydro tender is already delayed on land and site identification.
India has ordered a great deal of storage and built very little of it — about a tenth of what has been tendered. The prices are now workable and the targets are clear. What is missing is not money or technology: it is land, grid connections, and buyers whose credit a lender will accept.
What a risk manager should take from this
Four points that matter if you are pricing, hedging or lending against these assets.
Shape risk is now the dominant risk in a power book. A renewable generator sells at the average price of the hours it happens to generate, which is systematically below the period average and falling. Hedging with a flat baseload swap leaves that gap entirely open — and it is widening every year.
Volatility is the product, not the noise. Spain’s 596 negative hours and Germany’s €600 afternoons are the same phenomenon. Flexible assets earn from the spread between them, so an intermittency-heavy system makes flexibility more valuable, not less. Markets are already responding by shifting activity from day-ahead auctions toward intraday and balancing.
Correlation breaks the diversification assumption. Solar assets across one country are almost perfectly correlated at noon. A portfolio of ten solar farms in the same market is a concentrated position, not a diversified one. Real diversification requires different technologies, different weather regimes, or different timezones.
Storage revenue is a policy variable. Battery cash flow comes from energy arbitrage, ancillary services and capacity payments in a mix set by market design. Change the resource-adequacy rules, the ancillary product definitions or the viability-gap terms, and the asset reprices without a single physical thing changing. Model the regulation as carefully as the price.
The engineering question — can a grid run on mostly variable renewables? — is largely settled: yes, with enough flexibility. The open questions are who pays for that flexibility, through which market design, and on what timetable. Those are commercial and political questions, and they are where the risk now lives.
The engineering argument is settled — a grid can run mostly on wind and solar if you buy enough flexibility to go alongside it. What is not settled is who pays for that flexibility, and through which market. If you own or lend against these assets, remember that the revenue comes from rules as much as from physics, and rules can be rewritten without a single wire being moved.
Sources and further reading
- Electricity 2026 — Flexibility — International Energy Agency
- Solar and wind power curtailments are increasing in California — U.S. Energy Information Administration
- Europe’s record solar output drives surge in negative electricity prices — pv magazine
- EnergyWatts — India’s energy storage market, July 2026 — IEEFA
- Batteries and Secure Energy Transitions — International Energy Agency