A short and spicy post. There remains, even in 2023, a substantial fraction of the “future of energy” hivemind who are still convinced that the solution to all our problems is to build more transmission capacity, conveniently obstructed by the lack (so far) of the Act of Congress required to swiftly and forcibly appropriate the millions of acres necessary to string them across the country. NIMBYs, right!
[Edit Aug 2025: Austin Vernon and Ben Southwood at WiP write about the effect that batteries are already having on grid scale peak shaving, and their economic inevitability.]
In the background, of course, wind, solar and batteries have been continuing their steady, which is to say, explosive growth. Battery investment, manufacturing and growth are increasing by about 250% per year while costs continue to plummet. Indeed, while new solar farms take 5-20 years to pay for themselves, battery plants are so lucrative they’re often profitable by year two – which is unheard of in the energy infrastructure space!
So what? Here’s the key insight. Batteries and transmission are in direct competition. Both enable electricity arbitrage – the profitable repricing of a resource by matching different levels of supply and demand. Transmission moves power through space (technically null space, at the speed of light) and batteries move power through time. And while batteries have a fixed cost per MWh delivered (that is falling about 10% per year), transmission lines get more expensive as they get longer.
Intuitively, we should expect that for a given market, local energy generation landscape, demand profile, historical weather variability, etc, a grid storage battery would be competitive against a transmission line longer than a certain length, and this is true. The challenge for transmission is that as batteries get cheaper and NEPA lawsuits get more expensive, the competitive length for transmission lines is falling fast – the outcome is not in doubt.
Further stacking the deck in favor of batteries is the fact that power arbitrage depends on differences in demand, and there is a lot more spatial correlation than temporal correlation in energy demand. For example, over a 500 mile grid people will be using power for cooking and heating at the same times of day, while local weather systems will impact wind and solar generation in a correlated way. Conversely, power demand varies by a factor of two or three over a 24 hour cycle, every day, like clockwork. Why do batteries pay for themselves in 18 months while transmission lines, if built, lose money? Wonder no longer.
Let’s make this quantitative. I’ll reuse a couple of figures from “Geophysical constraints on the reliability of solar and wind power in the United States” by Shaner, Davis, Lewis, and Caldiera, published in 2017 (non paywalled version). Based on 39 years of historical demand and weather data, then prevalent battery prices suggested massive investment in grid upgrades was the only sure path to reliable, low carbon electricity supply in the US. Even at the time, a pragmatic evaluation of future trends and forcing functions in batteries would suggest otherwise, and by 2023, 6 years later, the data tells a very different story.
Fig. 2 Reliability of solar and wind generation as a function of area and resource mix. Contours and shading in each panel represent the average calculated reliability (% of total annual electrical demand met) by a mix of solar and wind resources ranging from 100% solar to 100% wind (y-axes) and aggregated over progressively larger areas of the contiguous U.S. (on x-axes compared to size of states (DC, NH, NY, CA) and regions (Western Electricity Coordinating Council, CONUS)). Storage and generation quantities are varied in each panel: (a) 1 generation, no storage; (b) 1 generation, 12 hours of storage; (c) 1.5 generation, no storage; (d) 1.5 generation, 12 hours of storage. These plots were generated by running each scenario for all 50 states, 8 NERC regions, and the contiguous U.S., respectively. For each resource mix simulated, the results were regressed (y = log(x) + b) and plotted as the shown heat maps. Thus, the plots represent the average area-dependence and effect of resource mix on ability to meet the total annual electricity demand in the contiguous U.S.; specific regions will be more, or less, reliable.
This chart allows us to compare the relative cost/benefit for different strategies, whether they be building out batteries, grid, solar, wind, or some mixture.
If we start at our present condition (red X), is it better to add more transmission (green arrow), add more storage (light blue arrow), or add more wind/solar (dark blue arrow)? Assuming the cost of each action is roughly equivalent, there is almost no benefit to adding more transmission, which after enormous expenditure of treasure and political capital would increase renewable penetration from ~55% to ~56%. In contrast, adding a bunch more wind and solar, but no storage would increase renewables to about 83% of total demand, reducing carbon emissions commensurately. Adding 12 hours of storage with no additional transmission or generation could increase reliability enough to support 86% of total demand, by far the biggest increase with the smallest impact on land use, as batteries are tiny compared to solar arrays, transmission lines, and wind farms.
Adding both solar and wind overproduction and batteries takes us into the lower right quadrant where a state-scale grid can support ~97% of demand, and once again adding (or subtracting) transmission (pink arrow) makes almost no difference.
Adding 12 hours of storage to the entire US grid would not happen overnight, but on current trends would cost around $500b and pay for itself within a few years. This is a shorter timescale than the required manufacturing ramp, meaning it could be entirely privately funded. By contrast, upgrading the US transmission grid could cost $7t over 20 years. This, incidentally, is why the future of electricity is local.
Fig. 3 Changes in reliability as a function of energy storage capacity (0–32 days) and generation. Lines in each panel show the reliability (% of demand met; x-axes, (a) linear scale, (b) log scale) of a mix of solar and wind resources aggregated over the contiguous U.S. and ranging from 100% solar (top panel) to 100% wind (bottom panel) as the installed generation quantity (left y-axis) or capacities (right y-axis) increase and the energy storage available increases (lines). Energy storage capacities of 0 and 12 hours, and 4 and 32 days are shown. In each case, the horizontal dashed line indicates the capacity at which total energy production and electricity demand over the 36 year period are equal (i.e. 1 generation).
To understand how renewables can nail down the last 1%, 0.1%, and so on, let’s look at the graphlets in the right column (log scale), that track how much overbuild is needed given some quantity of storage and generation mix as a function of desired reliability level.
In all cases, we see a grid without storage is useless. This should be intuitively obvious but it is good to see it confirmed here. Conversely, we see relatively small quantities of overall storage dramatically decrease the amount of generation overbuild needed. Over a wide range of wind/solar mix and battery installation, we can hit 99.99% reliability (significantly better than the pre-renewable grid) with < 10x nameplate overbuild. This is how the grid will look in the future.
These graphs assume a continent scale grid, but as we saw from the first figure, changing the size of the grid doesn’t affect reliability very much. Given a choice between 10x overbuild and 4 days of battery storage on some local DER or the equivalent reliability on a national scale with 8x overbuild and 3 days of storage, plus $10t of transmission lines that, on average, are mostly used to accumulate ice in ice storms or accrue deferred maintenance costs, it seems extremely clear to me which model will prevail. Adding marginal batteries until demand saturates will generate enormous profit and value growth. Adding additional grid is just stranding assets.
My views on this matter are unconventional, even controversial. Arguably this is my spiciest hot take on the future of energy. We won’t have to wait long to find out which model prevails. Bet against batteries, if you dare.
Addendum Jan 2024: While most grids are adequately backed up with just 5 hours of storage, due to diurnal demand variation, the question of solar+battery cost for other patterns of utilization comes up from time to time. I computed the minimum cost mix for any given utilization duty cycle and plotted the result as costs continue to fall over the next decade.
The takeaway from this chart is that power hungry high revenue growth loads, such as new AGI datacenters, which typically cost at least $50/W, will almost certainly prefer a beyond-the-grid solar+battery power supply as it is both cheaper and more readily available than nuclear, coal, or, in most cases, gas. The exception, for now, is new gas generation being codeveloped with a new datacenter somewhere with a bunch of stranded gas production, such as the Marcellus basin. Even then, I expect solar+battery costs will undercut gas, even for this application, in regions with plenty of gas, within a decade.
As expected, high energy consumption commodity industrial processes with low capex will always enjoy a cost advantage if they can adapt to intermittent operation, shown here by the knee in the curve at around 6 hours per day.
Addendum November 2024: Much more detailed calculations are performed in this recent post.