# Iterable Manufacturing Compounds Cost Declines Extraction Does Not ## Claim Learning-curve cost declines accrue to things manufactured repeatedly at scale and not to things dug out of a fixed physical stock, so the useful forecasting distinction is iterable production versus one-off extraction rather than clean versus dirty. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Applies to cost forecasting for any technology-versus-commodity comparison: solar panels and batteries behave like chips, fossil fuels behave like mining. - Predicts where forecasts will be wrong in a specific direction — modelers who treat a modular technology as a commodity will keep underestimating its deployment. - Large one-off construction (a nuclear plant, a bridge) sits on the extraction side of the line despite being built rather than dug, because each unit is bespoke. - Breaks down where a manufactured good depends on a genuinely scarce input; the learning curve then hits a materials floor. - This is about the shape of the cost curve, not about which energy source is preferable on other grounds. ## Evidence > [!quote] ID:rw_01m1sdh2cw25j1yegn0shawmpd > "as as resources decline. So, I think that's a key difference. Fossil fuels are commodities that you have to dig out the ground. solar when batteries are basically technologies where you can get these they're kind of like computing like some of the >> lots of predictions of solar power were incredibly wrong as you said the kind of hockey stick of hope >> and this was very very widespread like most people this took most people by surprise >> the people that I spoke to that didn't take it by surprise and they were kind of laughed at at the time were looking" Ritchie's account explains a forecasting failure rather than just naming it: the analysts who got solar right were the ones treating it as a manufactured technology on a learning curve, and they were dismissed at the time. The claim generalizes past energy — it is the same reason compute costs fell and bespoke infrastructure costs did not. The specific forecast-error history is recounted conversationally; the underlying literature on Wright's law is where to verify it. ## Links - [[3. Resources/MOCs/Product & Startup Instincts MOC]] — a cost-structure heuristic for deciding which curves a business can expect to ride - [[3. Resources/Permanent Notes/Zero Build Cost Shifts Advantage From Depth to Breadth]] — extends: both trace a strategic consequence from the direction a cost curve moves - [[3. Resources/Permanent Notes/Incremental Progress Is Invisible Because It Has No Headline Moment]] — from the same interview, and the reason the solar curve surprised people who were watching for an event rather than a trend ## Tests / falsifiers - Track a modular clean technology against a bespoke one over a decade. If both decline at similar rates, iterability is not the operative variable. - Find a commodity extracted from fixed stock that has sustained a learning-curve decline. That would break the dividing line as stated. ## Source - [[3. Resources/Readwise/Articles/Hannah Ritchie The Future Is Not Bleak]] — interview with Hannah Ritchie of Our World in Data. The learning-curve framing is standard energy economics; her contribution is applying it to why forecasters erred.