Thread by @ReflexFunds

Wright’s Law/Learning Rates/Experience Curves are extremely important to understand when considering technological progress & are particularly key to the Clean Energy transition & $TSLA.
However they are often misunderstood so I think it’s worth explaining some of my thoughts:

1) We have found consistently over the past 100 years that as cumulative historic production of a product doubles, the production cost reduces by a roughly fixed % (maybe 10-30%) - the “learning rate".

2) This is Wright’s Law and it is a rule of thumb which works over long time periods over multiple product and factory generations, but is not likely to be precisely accurate when making short term predictions.

3) Wright’s Law works because of three different underlying mechanisms:
A) R&D spend roughly scales with an industry’s revenue, so cumulative R&D scales with cumulative production. In a larger industry, more is spent on technological progress & breakthroughs drive down cost...

4) ... R&D appears to be the largest driver of experience curves but it doesn’t work on short time periods because new R&D could take several years to production. It can also be lumpy with cost breakthroughs coming in fits & starts when a new generation of tech is introduced.

5)
B) For every product produced the staff & company “learn” how to do their task better. This is both more productive staff & better methods introduced after fixing bottlenecks & problems previously experienced. In other words, all COGs of a product can double up as R&D.

6)
C) Increased economies of scale.
These are actually a function of annual production rather than cumulative production. Scale advantages could be higher staff productivity, fixed cost leverage or better purchasing power from suppliers (plus suppliers own scale savings)…

7) ... Economies of scale can actually cap out fairly early – you get huge benefit from 1 production line producing 500k cars vs 100k cars, but two 500k factories only has more limited increased economies of scale vs one 500k factory.

8) Its very important to note that Learning Rate is not fixed. It is possible to significantly accelerate the learning rate in your industry by: A) scaling up R&D as a % of an industry’s revenues or even just improving R&D productivity per $.

9)
B) Tightening the feedback loop between design, manufacturing and customer feedback with an agile development business model - this allows you to much more efficiently translate learnings from experience into better production and reduced costs

10)
C) Establishing superior engineering philosophies & corporate cultures: Focus on innovation, Physics first principles based mindset (break down difficult problems into achievable sub tasks), Use multi discipline knowledge & Reject expert's advice if based on bad assumptions.

11) It is also important to note there are some nuances to experience curve mechanisms:
A) Experience curves really operate on unique components – when looking at a system with multiple sub components you are really looking at an average experience curve over the whole product.

12)
B) Unique new components should follow very steep learning curves (because cumulative production volume is doubling very rapidly) while commoditised components should follow much slower learning curves (because it will take a long time to double cumulative volume).

13)
C) Commodity costs themselves might not reduce at all (or even increase in price if raw material production expansion is lagging behind end product capacity/demand).

14) For these reasons you can’t expect all 10,000 unique Model 3 components and processes to follow the same experience curves – some have already gone through 50 years of learning in ICE cars.

15) Most of an EV Powertrain should follow steep experience curves to some extent - this includes Battery Cells, Packs, BMS, chargers, converters, motors, inverters, cooling etc. But only for costs which are not for off the shelf components or commodities.

16) I’ll note here that most battery metal “commodities” are actually low historic volume, highly value add products with room for innovation which can themselves follow experience curves as EV sales grow – but this can also be offset in the short term by supply demand mismatch.

17) Many of the components supplied to Tesla are actually designed by Tesla in-house with just manufacturing outsourced – so many supplier components can also follow experience curves the same as components manufactured in-house.

18) Due to the tools Tesla uses to accelerate learning rates (higher R&D productivity, more agility to rapidly improve production processes, better engineering culture), Tesla can still achieve significant cost reductions in many of the car components common to both EVs and ICEs.

19) We are already seeing signs of this with what looks like a groundbreaking casting design for the Model Y body and a completely unique exoskeleton design for the CyberTruck body.

20) R&D in real manufacturing innovation for car production has been very low for decades as the global auto industry settled into comfortable local oligarchies with no culture of progress or innovation.

21) This means the industry has not been learning enough from the hundreds of millions of cars it has produced over recent decades and there is a lot of slack for Tesla to apply new technology to reenergise innovation in components which are already extremely high volume.

22) For this reason there is an argument that even non EV components in Tesla cars could follow relatively steep learning curves in line with Tesla’s cumulative production (rather than tracking much smaller increases in cumulative historical car volume from the past 100 years).

@threadreaderapp unroll

https://twitter.com/threadreaderapp/status/1223934256398983168?s=20