So now I am following up on BNEF battery pack cost prediction. BNEF predicts a pack cost of $94/kWh in 2024. Does this accord to basic statistical analysis? Let's see.
A Behind the Scenes Take on Lithium-ion Battery Prices | BloombergNEF
So BNEF is getting to their prediction through three steps.
1) Estimate the experience curve on cumulative battery volume. They get 18% price decline per doubling of volume.
2) Predict future pack size per vehicle.
3) Predict future EV sales and compute cumulative pack volume.
Steps 2 and 3 are opportunities for the analysis to become poorly calibrated and for modeler bias to creep in. I am taking a shorter route.
1) Estimate experience curve based on cumulative PEVs sold. Pack cost declines 28% per doubling of cumulative PEVs.
2) Forecast cumulative PEVs to be sold.
Notice that I don't need to fiddle around with making pack size assumptions, nor is that even appropriate. Cell cost is about 2/3 of total pack cost, but the pack specific costs may appropriately be thought of as one pack as one unit. Also we should expect the capacity per pack to increase as the cost per kWh declines. So my approach will tend to capture that. Another advantage of indexing the learning curve on EVs sold is that is it fundamentally EV sales that will be motivating R&D and scale up investments.
A critical difference here is that BNEF obtains a learning rate of 18% cost reduction per doubling of kWh volume, but my rate is a reduction of 28% per doubling of cumulative EV sales. So if one were to assume that kWh/EV where to stop growing, you might conclude with BNEF that the pack cost won't come down as fast as my model suggests. So that makes BNEF assumptions for step 2 critical.
In both modeling cases, though, we need to fight the temptation of thinking that mineral costs set a firm limit to how cheap packs can become. This would only be true if mass and mineral composition of a kWh of capacity were somehow fixed. But with increases in energy density and changes in chemistry, this is simply not true. Part of the experience curve is advancing the chemistry of battery packs finding more dense and low cost options along the way.
What will slow down the decline of pack costs will the size of the EV market. As EVs saturate the market, they shift to growing a single digit rates, so roughly linear growth. So this make it very hard to keep doubling. Fortunately this is not much of a problem until the 2030s.
So what about BNEF's 2024 prediction? My model suggest that $100/kWh will be obtained between about 10M to 50M PEVs made (regardless of how long it takes to get there, so my step 2 is not critical.) But assuming my mean logistic forecast in other posts and that the auto market grows about 3% per year, with 95% confidence the $100 threshold will be breached between 2020 and 2023, not as late as 2024. In deed for 2024, I've got a confidence interval from $35 to $78 per kWh. Now as with any predictive statistical model, if conditions fundamentally depart from what has been experienced within the training history, these projections can fail. This is basically the no-black-swans caveat.
Also in terms of when Peak oil comes, sub-$100/kWh come by 50M cumulative EV sales and oil peaks by about 75M EV sales. So these events seem likely to come just one or two years apart. I still do not see $100 as some magic number when the planets align to make EVs a thing. The fact is every year there is a double digit drop in cost, there will be an increase in EV sales. Moreover, there are many other critical elements to making EVs more powerful and cheaper to build, like improvement in motors, inverters, software control, and more. All of these components also have their own experience curve. So year after year EVs will be increasingly more desirable, more available and more affordable. Singling out the battery pack as having a make or break price risks losing sight of all the many things that advance EV adoption along the way.
One other little point, at times I've been critical of how BNEF uses a volume weighted average. Tesla appears to be leading the pack. But really this does not matter. Basically, producers that are able to get the cost down lower will tend to grow their sales faster than others. The laggards lose share of volume along the way. Part of what will continue to sustain progress down the experience curve is competition weeding the overpriced packs out of the market. If BNEF were tracking the minimum cost instead of the average cost, we would have more worries about jumps in the technology rather than steady progress to lower prices. So volume weighted averages are just fine by me.