The biggest energy story of the last fifteen years is the rise of solar photovoltaics, also known as solar PV or simply solar panels. Solar PV was invented in the 1950s, and began to be used in appreciable volumes for utility-scale electricity generation in the US in the early 2000s, but only around the 2010s did it start to become a large share of planned generation projects worldwide.

Since then, solar generation capacity has grown incredibly quickly. By some metrics, solar PV has been deployed faster than any other energy source in history, going from 100 terawatt-hours of generation to 1,000 terawatt-hours in just 8 years, compared to 12 years for wind and nuclear, 28 for natural gas, and 32 for coal. In the US, solar PV projects are by far the largest share of planned new electrical generation capacity. Of the roughly 1,900 gigawatts of electricity generation projects in the interconnection queue, around 50% of them are some type of solar PV project.

But while solar PV is growing rapidly, in absolute terms it’s still fairly small potatoes. As of 2023, solar made up around 4% of overall electricity generation, and less than 1% of total US energy production. That means that the main questions around solar PV are about its potential: how long can its rapid growth rate continue, and how large a fraction of our energy can it effectively supply?

The answer to this question is shaped by two salient facts about solar power. First, the cost of it has fallen precipitously over time. Since its invention in the 1950s, the cost of solar PV has fallen by a factor of close to 10,000. In the last 10 years alone, the cost of solar PV cells has fallen by more than 50%, and they’re projected to get even cheaper. This has made solar PV one of the cheapest methods of electricity generation.

The second salient fact about solar PV is that it can’t generate electricity on demand. Unlike technologies that generate energy by burning fuel which can be turned off and on as needed (such as gas, coal, nuclear), solar is intermittent, and only generates power when the sun is shining.

The future potential of solar power is, broadly, a function of these two factors. Some folks think that solar’s intermittency will fundamentally limit how much of our energy it can supply. Therefore, they believe, we should deemphasize solar in favor of “firm” sources of energy like gas turbines, next-generation nuclear or advanced geothermal. Others argue that solar PV and storage batteries will get so cheap that its intermittency will become less and less of a factor: the cheaper it is to build, the more you can address solar’s intermittency by simply building more panels and storage to compensate.

While it’s hard to predict the future, some simple modeling seems to favor the second outcome. Supplying a large fraction of energy consumption purely on solar power does indeed require a large degree of “overbuilding”: that is, building solar PV and storage capacity greatly in excess of day to day energy consumption. But even with this overbuilding, it won’t take much progress in solar and storage costs for solar to be as cheap or cheaper than today’s electricity, even at very large fractions of electricity consumption.

Solar panels work by converting light from the sun into electricity. A slightly more detailed explanation is available at my previous essay about solar power, but briefly, solar panels consist of semiconductor components called p-n junctions. When light strikes the p-n junction, it excites electrons, which get pushed to one side of the p-n junction by an electric field within it, leaving electron holes on the other side. Connect the two sides of the junction via wire, and electricity will flow as the electrons try to reestablish equilibrium.

Different types of solar panels will have different conversion efficiencies (the fraction of solar energy converted into electrical energy), but utility-scale panels in the US are generally 20–23% efficient. The more intense the light, the more power the panel will produce. On Earth, sunlight reaches the top of the atmosphere with an irradiance of 1,360 watts per square meter, but this gets attenuated as it travels through the air, and at Earth’s surface irradiance is about 1,000 watts (1 kilowatt) per square meter when the sun is directly overhead and not blocked by clouds. So a 21% efficient solar panel will have a maximum output of 210 watts per square meter.

For a given spot on the Earth’s surface, irradiance will start at 0 watts per square meter when it’s completely dark, gradually rise to its maximum, and then fall back to zero at night. The graph below shows solar irradiance at the equator during a day in March.

Solar irradiance is maximized when light is perpendicular to the surface being struck: for a horizontal surface on the ground, this occurs when the sun is directly overhead. Because the earth is approximately a sphere, the farther north you go, the more tilted the surface of the earth will be in relation to the sun (i.e: the lower the sun will be in the sky), reducing the irradiance on a horizontal surface. And because the tilt of the earth’s axis changes in relation to the sun during the year, solar irradiance will also change depending on the time of year: in the northern hemisphere, irradiance rises in the summer and falls in the winter. Below is the solar irradiance in Atlanta, Georgia (where I live) compared to the irradiance at the equator on the same day in March.

And here’s the maximum irradiance on a horizontal surface in Atlanta over the course of a year.

This means that the same solar panel in the same location will produce markedly different amounts of power at different times of the year, even before considering things like cloud obstruction. If I have 20 square meters of solar panels with 20% conversion efficiency mounted flat on the ground near Atlanta, on a cloudless day in June these panels will generate close to four kilowatts of power at their peak output. On a cloudless day in January, their peak output will be around half that.

Day-night cycles, changing solar irradiance depending on time of year and geographic location, and cloud obstruction all mean that solar PV panels are often producing much less power than they theoretically could. The average capacity factor of utility-scale solar PV in the US is around 23%, meaning that on average they produce 23% of the power they would if they were exposed to 1,000 watts per square meter of sunlight 24 hours a day. This capacity factor varies by location, with sunny Southwestern states having higher capacity factors than Northeastern states.

Peak power generation for a solar PV system will be in the middle of the day, when the sun is highest in the sky. This doesn’t align particularly well with patterns of electricity consumption, which tends to be highest in the early evening. For our 20 square meters of solar panels in Atlanta, here’s power produced, superimposed over the power consumption for a single family home.

We can see that the four kilowatts of power is substantially more than peak demand, which is less than two kilowatts. For most daylight hours, our solar system can more than meet our power demands. But peak demand occurs in the early evening, when solar production is virtually nothing. And our simulated house continues consuming power all through the night, when there’s no sunlight at all. This misalignment between electricity demand and solar PV supply is what produces the famous “duck curve,” which shows power required from non-solar sources when there’s a significant amount of solar capacity on the grid.

In winter, the situation is even worse. Our simulated Atlanta house uses a heat pump for heating, driving up wintertime electricity consumption. Our January peak power demand is now more than 9 kilowatts, far more than our solar PV system can produce even at its peak. And reduced solar irradiance means that peak power production is down to around two kilowatts.

In June, our 20 square meters of solar panels could supply about 52% of our home’s electricity demand. In January, it could only supply about 10%.

And in fact, the situation is even worse than this, because this simple simulation doesn’t include cloud cover. Not only will clouds sporadically reduce the power generated from our panels, but cloud cover tends to be higher in winter, further reducing our already-anemic wintertime PV output. Here’s a week of output from our simulated 20 square meter PV system in January, with both no clouds and a simple cloud simulation.

This intermittency is the fundamental challenge with solar energy. We can’t turn sunlight off and on at will, and the amount of light a solar PV panel receives will vary depending on the time of day, time of year, and where on the Earth that panel is placed. At the same time, electricity demand will vary significantly, both over the course of a single day and seasonally, in ways that aren’t particularly convenient for solar PV. Peak electricity demand happens well outside peak solar hours, both over the course of a day and (assuming electrified heating) over the course of a year.

There are a few different ways we can address this intermittency problem. The most obvious one is to just use other sources of power when the sun isn’t shining; either power sources that can be turned on and off on demand (such as gas turbines), or other intermittent sources whose peaks are offset from solar (such as wind).

If we want to use an exclusively solar system, we have basically two strategies for dealing with intermittency. One is to increase the amount of power generated from our solar panels: this won’t let you generate power at night (on earth, anyway), but enough panel capacity can allow you to meet power demands in winter, low-light morning hours, or when it’s cloudy out. There are a variety of ways to increase panel power output:

• You can build more solar panels.

• You can use solar cells with a higher conversion efficiency. Some research cells have reached nearly 50% efficiency, and the average conversion efficiency of commercial cells has trended upwards over time.

• You can tilt the solar panels to face the sun at a more favorable angle. Tracking solar systems mount motors to the solar panels to allow them to follow the sun. In the US, roughly 96% of new utility-scale systems have single axis tracking (ie: the panels can vary their tilt in one direction). And even non-tracking utility-scale systems will typically be built at a fixed angle chosen to maximize sunlight exposure based on geographic location.

• You can use mirrors or lenses to concentrate more light onto your solar panels. This type of system is called concentrator photovoltaics, but it’s not particularly popular.

• You can put solar panels into space, in an orbit that lets them receive sunlight nearly 24 hours a day. This is known as space-based solar power. Historically the costs and difficulties of launching panels into space and transmitting the energy back to earth have made this type of solar largely theoretical, but the idea of 24/7 solar is enticing and there’s always someone trying to make it happen.

The other way to deal with intermittency is by adding some form of energy storage. There are a variety of ways that this could be done — using electricity from our PV panels to spin a flywheel, or pump water uphill, or synthesize methane to later burn — but in practice today most new energy storage is in the form of large lithium-ion batteries.

In practice, dealing with intermittency requires both increasing the power produced by our panels and adding storage. Without storage, the biggest ground-based solar array in the world still produces zero power at night, and storage is only useful if our PV system is large enough to produce excess energy beyond what’s needed for immediate consumption.

What’s the relative value of additional solar PV capacity vs. additional storage? We can understand this by running simulations with different amounts of solar panel capacity and energy storage, and comparing it to the power demand for our theoretical single family home. The graph below shows power generated for our single family home at different solar PV capacities, without any storage.

With solar panels alone, we max out at slightly less than 50% of annual power demand met by our solar PV system: no matter how many panels we add, our system still can’t provide any power at night. But let’s see what it looks like when we start adding storage capacity.

As we increase the amount of storage, we can supply greater and greater proportions of our household’s electricity demand, reaching over 99% with 42 kilowatts (~200 square meters) of PV capacity and 80 kilowatt-hours of storage. This is around four times our maximum household power consumption, and roughly 40% more storage than the capacity of a base Tesla Model 3.

But both our storage capacity and our solar PV capacity are subject to diminishing returns. Each additional square meter of panel and kilowatt-hour of storage supplies a smaller and smaller proportion of our overall electricity demand, meaning each additional kilowatt-hour of electricity provided is more expensive than the last. We can supply 25% of our household’s electricity demands with 15 square meters of panels. But supplying 75% requires not three times the panel area (45 meters) but more than five times (85 meters), plus 20 kilowatt-hours of energy storage. And going from meeting 75% of demand to 95% of demand requires roughly doubling the size of the system, to 135 square meters of panels and 50 kilowatt-hours of storage.

So, while solar can easily supply relatively small fractions of electricity demand, this gets harder and harder, requiring more and more infrastructure, the greater the fraction of our energy we want solar to supply. For our simulated household, meeting 100% electricity demand requires 52 kilowatts (about 250 square meters) worth of panels, roughly five times our peak yearly consumption, and enough storage to power our house for almost four days straight during the summer. And even this might not be enough if we got a bad string of cloudy winter days. (The exact size of the system needed to meet ~100% demand can vary, with fewer panels trading off against more storage to some extent, but the basic point remains.)

But if solar and storage gets cheap enough, this “overbuilding” becomes much less of a concern: low enough costs mean that the system can remain competitive even as each additional kilowatt-hour gets more expensive than the last. When buying a new computer, no one worries about wasting money on excess hard drive space that only rarely gets used: hard drive capacity is so inexpensive we can just buy enormous quantities of it to cover whatever our future needs might be. So understanding the potential of solar PV means understanding what its cost trajectory looks like.

The costs of a solar PV system can be broken down into hard costs, soft costs, and operation and maintenance (O&M) costs. Hard costs are the costs of the physical solar system itself: the photovoltaic panels, the inverters, the racking, the labor required to install it, and so on. Soft costs are design work, permitting, developer overhead, and other things not directly associated with building the physical system. O&M costs are those incurred during the course of operation, including insurance, land leases, cleaning, replacing damaged parts, and so on. Below is a current hard and soft cost breakdown for fixed-tilt utility-scale solar in the US.

Overall US costs are slightly more than $1,000 per kilowatt. We see that thanks to 70 years of learning curve improvements, the solar PV cells themselves are less than 1/3rd the cost of the overall system. The shrinking fraction of the cost of PV cells vs the rest of the system are why there’s interest in things like ground-mounted solar which can eliminate racking entirely, and reducing installation costs by robotically installing solar panels.

With storage, we see a broadly similar pattern. The costs of utility-scale storage can be broken down into the hard costs of the system itself, various soft costs associated with installing it, and ongoing maintenance costs. As with solar PV, the costs of lithium-ion batteries have steadily fallen over time due to various learning curve effects (increasing scale, more materially efficient batteries, and so on), though the costs of the battery itself remains a larger fraction of overall storage system cost than the module is for solar PV.

These are capital costs, but what about operations and maintenance (O&M) costs? Because solar and storage systems don’t require purchasing fuel, and have almost no moving parts, operations and maintenance costs are low. NREL estimates that for utility PV, O&M is about $16 dollars per kilowatt per year, or about 1.5% of capital costs annually.

For maintenance costs for battery storage, NREL doesn’t give a line-item breakdown, but estimates the annual cost of O&M for battery storage is about 2.5% of the capital cost, or around $12 per kilowatt-hour for current US battery costs.

We can put this all together to calculate the levelized cost of electricity (LCOE) for our solar plus storage system and different fractions of electricity supplied. LCOE is simply the cost of an electric power generation system, divided by the amount of electricity it produces, and adjusted by the discount rate (which reduces both the present-day costs of future payments and the present-day value of future electricity generated).

(In reality, it would not be possible to build a small-scale solar system for these costs; residential solar PV costs are much higher. The purpose is to illustrate how costs vary as solar PV serves a larger fraction of electricity demand. Though we’re simulating a single home, the same principles apply to grid-scale demand.)

Below is a graph showing LCOE for the different combinations of solar PV capacity and energy storage we looked at earlier. This is based on the following parameters, which are roughly current for US utility-scale solar:

• Solar panel cost: $1100 per kilowatt, $17 per kw per year O&M

• Battery cost: $476 per kilowatt-hour, $12 per kwh per year O&M

• Solar panel lifetime: 35 years

• Battery lifetime: 20 years

• Solar panel capacity factor: ~20% (varies depending on cloud simulation)

• Solar cell efficiency: 21%

• Discount rate: 7%

We can see that without any sort of storage, and with low amounts of solar (where the power can simply be used immediately without any going to waste), our solar system costs around 5.7 cents per kilowatt-hour. This is smack dab in the middle of what Lazard lists as the current range for LCOE for utility-scale solar in the US, and slightly more than the average LCOE for recently built US utility-scale solar plants.

However, as we expand the size of our system to serve a greater fraction of our electricity demand, our cost per kilowatt-hour quickly rises. At 50% of electricity served, we’re at 13 cents a kilowatt-hour. At 70% we’re over 16 cents. At 90%, we’re nearly 25 cents.

How much does this change if solar and storage costs continue to fall? The below graph shows electricity costs at current US costs for solar and storage, compared to the current European costs ($750 per kW for PV, $300 per kWh for storage), and potentially future achievable costs ($400 per kw for solar, and $150 kWh for storage).

Even just driving down costs to the point of the current efficient frontier would cut the costs of US solar by roughly a third, allowing for costs less than 10 cents a kilowatt hour in our model even at 50% of electricity supplied. If costs could be driven down even further, to what experts project we might reach (BloombergNEF projects a future cost of $400 per kilowatt for solar), we could reach nearly 80% of electricity demand with solar and storage for 10 cents a kilowatt hour, roughly the average cost for residential electricity generation in the US today.

Of course, there’s no guarantee that solar costs will continue to fall. Even if the learning curves for batteries and solar cells continue (and they might not: learning curves work until they don’t), solar costs might stay flat or even increase if we don’t find ways to drive down the costs of other parts of the system. We’re already seeing this to some extent. Per the Solar Energy Industry Association, utility-scale solar costs in the US have been roughly flat for the last six years, even as module costs have fallen, due to rising costs elsewhere in the system.

The main advantage of solar, aside from environmental considerations, is its low cost. A solar cell requires no fuel, has no moving parts, can be produced in large-scale factories in enormous volumes, and requires very little maintenance. Lay it down on the ground and it will start generating electric power, no muss no fuss.

The main drawback of solar is that we can’t control when the sun shines: day/night cycles, clouds, and seasonal changes in the sun’s position all reduce how much our solar panel generates, and make it hard to generate a given amount of electricity reliably. The greater the share of electricity supplied by solar power, the more extensive the infrastructure required to overcome this drawback, and the higher the cost.

But as the costs of solar continue to fall, this drawback becomes less binding, and it becomes economically justifiable to “overbuild” solar infrastructure and use it to serve increasingly large fractions of our energy demand.

Thanks to Austin Vernon for reading a draft of this post. All errors are my own.