A detailed exploration into what drove the madness in the battery storage procurement and origination markets over the last three years… and what helped us get back on track – by Dan Vickery
I. Roadmap of this Paper
In a previous role in June 2020, I purchased hardware and EPC for a 100 MW / 400 MWh storage project and then took a two-year hiatus from the procurement game. I’d heard rumblings of storage price increases over that period (who wasn’t hearing about inflation in 2021?), but I was stunned when I received a bid for a storage project exactly two years later. In those two years that I was blissfully ignorant of what was happening, the turnkey constructed cost of a utility-scale energy storage system increased 40%. This paper explores what happened.
First, this paper breaks down that 40% turnkey price increase by building up the turnkey price of a theoretical 100 MW / 400 MWh energy storage project in each month between June 2020 and June 2023. In addition to being a point at which I had good visibility into pricing, this starting point is also relevant because the industry had finally settled on broadly using (a) LFP liquid cooled batteries and (b) modular form-factor containers – and thus general consistency allowing apples-to-apples comparisons.
Next, it shows the positive impact that the learning curve had over this period to help lessen the pain of the pricing shocks that had occurred.
Then, using the monthly CapEx buildup as an input, this paper presents what a monthly curve of where toll pricing transacted in this period (or, broadly, where pricing should have transacted)… absent the stand-alone storage ITC. And then it shows the same toll pricing curve with the addition of the stand-alone storage ITC, and highlights the significant impact that the ITC had for strike prices.
Last, we weave that journey together to show where we are today (spoiler alert: we’re almost back to exactly where we started) and theorize some paths forward.
But before you dive in, I think it’s important to caveat this analysis a bit. I believe the methodology is defensibly sound – I had an “aha” moment when my model seemed to be reasonably calibrated to vendor quotes received over different periods, and I also passed this by a few industry friends who generally agreed with my logic and trends they saw. In any case, this is a necessarily narrow-scoped exercise that requires limiting variables to the extent possible (including, among others, being vendor agnostic) to obtain any kind of insights. The price buildup attempts to capture middle-of-the-road pricing for the most common project designed in this period – one that uses modular equipment with China-sourced LFP batteries. And the same goes for the toll pricing analysis: there are so many variables that will determine what price actually gets contracted for a given project (toll tenor (I assume 15 here), sponsor discount rate, sales and property taxes by locale, interconnection cost, volume agreements, debt terms, merchant tail assumptions, I could go on…) – so again, I sought assumptions where you could reasonably project tolls to be at the middle of the bell curve (full well knowing that tolls certainly would be struck above and below these numbers depending on exact circumstances). My point is: if you’re a utility that just signed a toll and you look at these numbers and think “I got screwed” or an asset owner that says “he’s dreaming” – just take a breath and consider what this is broadly trying to accomplish. Unless of course you just signed a $40 toll, in which case, yea, you got screwed.
When I opened that first spit-out-your-coffee-high quote in June 2022, I reflexively (a) was offended and (b) believed they were sandbagging, but then (c) started to connect with peers and understand what was actually happening. My thought process went as follows: “COVID, supply chain bottlenecks, inflation – OK, so all that is… maybe 10–15%, but shouldn’t learning rates at least partially offset that, so worst-case we’re looking at a few percentage points higher on build price than in a pre-COVID world?” Man was that simplistic. Let’s dig in.
The underlying drivers are presented in the chart below. As shown, inflation across the economy has paled in comparison to the inflation that the energy storage industry has experienced the last two years; in fact, looking at the chart below, it appears as though the Consumer Price Index (CPI) and Producer Price Index (PPI) were unchanged in that period! The two pieces clearly sticking out here are shipping costs (a temporary 600% increase) and raw lithium carbonate (peaking at a 1400% increase that has since settled down into the 400–800% range). The CPI, the PPI, steel, and copper had “insignificant” increases of 20%, 40%, 60%, and 70% respectively.
So how do all of these pieces overlap and impact the cost of a turnkey constructed and operating energy storage power plant in the US? To answer this, I built a model that priced out a 100 MW / 400 MWh facility for each month between June 2020 and June 2023. That model starts with granular pricing inputs and then escalates each individual cost forward by a cost multiplier that is rooted in the underlying inflation metric / commodity curve applicable to that cost component, but also included de-escalators for battery density improvements and manufacturing scale/learning rates (more on that later). The breakdown, cost components, and escalator applied to each individual cost component are shown in the table below.
Turnkey Plant Cost Component (left column) and Escalation Factors Applied (right column)
What those up, down, and sideways escalators operating in tandem imply for the price of a battery storage project “built” between June 2020 and June 2023 is best translated into the chart below. I place “built” in quotes because this analysis provides a snapshot view of the turnkey cost component buildup in each month – meaning the cost breakdown for a plant if it were to assume each component was purchased / locked in at exactly the underlying cost factors in that month. This view does not take forward or backward averages (which is indeed how vendors index their products); does not consider time gaps between contracting, manufacturing, delivery, and construction; and does not factor in volume agreements or component pricing risk either. When looking at the chart, three themes clearly emerge (as previewed in the underlying cost factors chart shown earlier):
- First, raw lithium carbonate had a staggering impact. Lithium goes from an almost inconsequential piece of the cost stack to about 20% of the equipment and 16% of the all-in constructed price (or greater when you factor in tariffs and sales tax increases associated with the finished goods); three years ago, lithium carbonate was essentially a rounding error (1.5% of the turnkey cost). This lithium price shock has led most vendors to now “index” quotes based on the price of raw lithium, which in and of itself created yet another challenge for buyers.
- Second, shipping (and trucking) costs swelling in the Dec. 20 to Dec. 22 timeframe indeed had a very dramatic impact. These are large, bulky products that primarily ship as one unit per container; whereas the marginal cost to a consumer good like an iPhone might be negligible, there are very real consequences here.
- And third, all other increases in underlying costs amount to almost no change in the overall cost of an energy storage project! Or, said another way, the wild ride we’ve been on can almost entirely be attributed to those first two components.
The chart below shows the cost buildup in $/MWh Nameplate of a 100 MW / 400 MWh storage plant “built” between June 2020 and June 2023.
One other important note: this was a purely cost-based exercise that did not / could not capture two important pieces happening in 2021. The first of those is the extreme shock to battery supply that was driven by battery-related factory shutdowns but also by delays / challenges in getting sub-components – the same issues that drove car manufacturing shortages globally (e.g., computer chips). The second was the simultaneous ramping of the EV industry – where nearly all available supply was prioritized and committed to the vehicle market. As I understand it, this was less a function of cost and more a function of whether you could even get your hands on a battery. In any event, that would mean the chart above likely underestimates actual costs paid (or glosses over the fact that you simply couldn’t find a battery supplier at that cost) for the better part of 2021.
If a rising tide lifts all boats… battery manufacturing scale and density improvements are the rising tides that kept a sinking boat from hitting the ocean floor. The learning curve is a bit difficult to capture in real-time (and takes a substantially more nuanced effort than the one I’ve done here), but I believe there are quite noticeable (and historically consistent) improvements that have happened in this period. In this analysis, I noticed what amounted to a 3% annual battery pack density improvement (aligning to historical trends), a 7.5% manufacturing scale / learning rate improvement, and a 2.5% EPC learning rate. These pieces play out in the real world less gradually and more as step-changes (large factories come online, denser chemistries hit the market, etc.), but in the long run, the trend should hold.
The “learning rate” is best understood as the reduction (as a percentage) in the unit price of a module as a function of the number of units shipped (of course as a result of the learnings therein). The learning rate for solar PV has been consistent and exceptional: with each doubling of solar PV module shipments globally, the price of modules decreased by 24%. Over and over again – since 1976 – this rate has held remarkably steady. If we parallel this metric to the storage industry – our perceived learning rate of 7.5% annual is reasonable if not conservative. BNEF indicates 3.5x growth over our analysis period – approximately 20 GWh of global installations in 2020 and 70 GWh in 2023. Thus, we should expect the learning rate over a 3.5x growth period to exceed the 21% cumulative learning rate used over the three-year period.
But even with our presumably conservative learning rate, the impact of the learning curve has been notable. Without any kind of learning curve or other improvements, I’d suspect we would have seen pricing closer to the $450/kWh all-in mark… and staying around the $400 mark still today. Thus, pricing coming down from the highs of February 2022 can quite meaningfully be attributed to battery density, chemistry changes, innovation, and manufacturing scale / improvements nearly as much as it can been attributed to reductions in lithium carbonate commodity pricing. See the chart below with a “counterfactual case” that reverses (adds back) the impact of the learning curve.
But let’s zoom out from a woe-is-me, lithium-is-expensive mindset. One of the silver linings of the pain we endured is that this price shock likely torqued the screws for driving manufacturers to obsessively focus on improvements and costs reductions for the long-term health of their business (and in a way, the industry) much more critically than prior efforts. Tesla, for example, just broke ground on a lithium refinery in Texas (to better control cost and its own destiny) and almost every manufacturer out there is increasing its R&D focus on alternative chemistry combinations or other step-change improvements to reduce battery pack costs. Long-term, this efficiency-driven focus nets the industry to a better position overall, but what exactly does this mean for what’s next? I’ll pick that conversation back up in Section VI below.
Storage CapEx offers little usable information until translated into a metric that yields whether a project will pencil, and more holistically, whether a storage project reaches price parity in a competitive power market. To do that, I ran the CapEx pricing for each month through a project finance model to yield a 15–year toll strike price (contracted revenue) based on that CapEx. To isolate the impact of the changing CapEx, the model held everything constant except for the CapEx and anything that was tied to CapEx (e.g., property tax as a percentage of CapEx). The resulting toll price shows us – based on each CapEx assumption – the price at which a sponsor would execute a toll. And taking that a step further, we can then assess whether that number can compete in the various North American ISOs.
The chart below shows the required 15–year toll strike price that a project sponsor would require to construct and operate a storage project based on the time-varying CapEx rates provided above assuming (a) June 2020 interest rates and (b) the then-applicable interest rate. Using figures in the chart below to explain: an asset owner would need to sign a 15–year toll of at least $22.66/kW-month in order to finance and construct a plant assuming January 2023’s CapEx, but only a $14.90/kW-month toll assuming June 2020’s CapEx.
Funny how we thought storage and toll pricing was high in mid–2020, huh? Since then, and shown above, toll pricing at one point in time had increased about 50% above the June 2020 mark. Interest rates have had an almost equivalent impact on toll pricing to the impact that surging CapEx (peaking in Feb 2022) caused. If we held interest rates constant to June 2020’s figures (with underlying near-zero Fed Funds Rate levels), we’d be looking at toll pricing today that is a mere 15% higher than three years ago; instead, we’re still up 44% versus that June 2020 mark!
Or are we…?
No, we’re not, because the stand-alone storage ITC provided a timely and long-overdue policy tailwind.
The stand-alone storage ITC in the Inflation Reduction Act (IRA) could not have come at a more critical time. The graph below replicates the chart in Section IV above but adds 30%, 40%, and 50% ITC scenarios. Where recent storage price shocks effectively priced out storage from almost every power market in the US (outside of the likes of the CAISOs that have almost no other option for capacity resources), a recent subsidence in lithium pricing and the passage of stand-alone storage ITC brings us close to historical “normal times” parity. Today with the ITC, we are somewhere around 6% higher (30% ITC) to 16% lower (50% ITC) than June 2020’s pricing environment that was characterized by low lithium pricing, low interest rates, and no storage ITC. That’s exciting, particularly given the industry’s sentiment just 12 months ago – and is particularly exciting to consider where the storage industry is headed as we (hopefully) soon transition further into this “normal” pricing world.
So, to summarize the above: insane lithium pricing and elevated interest rates ratcheted storage pricing through the roof and priced projects out of almost every power market in the country, but the trusty learning curve and the stand-alone storage ITC together have set storage essentially back to where it was three years ago. Storage is again reaching and exceeding thresholds that it needs in order to achieve price parity (and compete) in wholesale power markets throughout the US.
But where do we go from here? How do we think about all these drivers and project them forward? Surely, we’re beyond the days that our nation’s preeminent renewable supply chain managers can rely on their proprietary, highly strategic, never-fail approach of “slap a 15% de-escalator on it”, but since that’s what we do best as an industry, let’s do it anyway – but also incorporate the learnings above.
To assess where costs might be 12 months from now: let’s carry forward the same 3% density improvement and 7.5% manufacturing improvement that seemed to hold true the last three years; assume increasing mining operations bring down lithium “moderately” to 200,000 CNY/T (versus today of around 300,000); and assume risk-free interest rates come down by 100 bps. I don’t think that package of assumptions leans overly conservative, but I’d be hard pressed to be convinced that they’re aggressive. That puts toll pricing next year right around the $14/kW-month price mark – a near 12% reduction from “today’s” pricing and a price that tilts the scale toward price parity in increasingly more markets. And to project further into the future, we can use ITRPV long-term learning rates and BNEF installation forecasts. BNEF expects installations globally will double (above 2023) by 2025 and again by 2028 – thus achieving a 24% cumulative learning rate twice by 2028. Ignoring any other efficiencies and improvements that may happen in this time, we should expect toll pricing in the range of $13/kW-month by 2025 and $12/kW-month by 2028.
From that perspective, it’s quite comforting to think that an industry that was just about to hit its stride three years ago is getting back on track and might just be gaining its swagger again… and it doesn’t take huge leaps of faith to believe that.
But not if lithium spikes again.
 I use “cost” and “price” interchangeably throughout this paper. Both are intended to mean the all-in price (including any markups, etc.) paid by a project sponsor to fully procure and construct a storage project.
 I use the term “lithium” and “lithium carbonate” interchangeably in this piece. Both are intended to mean lithium carbonate, the commodity input to an LFP battery that has had the most impact recently to pricing and of which ESS vendors use to index their quotes.
 $/kWh nameplate presents a 110 MW / 450 MWh installed system for a 100 MW / 400 MWh “nameplate” project (a term often interchangeable to the term “usable energy”). So, this takes the cost of a 450 MWh installed system and then divides by the 400 MWh nameplate rating (thus yielding a higher $/kWh price than if it were divided by the actual installed capacity). Storage people always talk past each other with “$/kWh”, so I thought it might help to clarify a bit here.
 International Technology Roadmap for Photovoltaic (ITRPV) Results 2022. Fourteenth Edition, April 2023.
 In fairness, the lion’s share of this effort is dedicated toward reducing costs of EVs – but has the same benefits to energy storage.
 I initially tried to isolate this analysis only to CapEx changes over time (and thus ignore the impact of interest rates)… but clearly, they’re just too significant to ignore entirely. My approach takes a simple “add 250 bps above each applicable SOFR” (or “SOFR + 250”, in project finance parlance). This might be a slight oversimplification – as Dave Riester highlighted in Segue’s most recent paper Clean Energy vs. Recession, there wouldn’t be a perfect 1:1 movement in the applicable project interest rate with a movement in the underlying risk-free rate – but this provides a sufficient proxy for analysis. Plus, it is probably slightly truer for storage than for solar anyhow, given the perceived risk profile of the still-nascent technology as compared to solar.