I love these cats. Price transparency and efficiency in this market is waaay overdue. I would have never have guessed that chip prices rise over time. Seems like on the very lowest level of chip complexity, Moore's law doesnt apply.
Think supply and demand rather than Moore's Law. If the supply of a chip goes down (e.g., a manufacturing plant gets hit by an earthquake) or demand goes up (Apple designs the chip into the next version of the iPhone), then its price could definitely go up.
I'd guess that the markets for chips would be driven by similar dynamics as other commodities, such as oil.
Maybe the next step will be real-time feeds for chip prices, and after that people will start trading chip futures on Wall Street?
Economies of scale also play into this pricing. You can have chips whose demand is actually quite low but enough for there to be some market and they could increase in value relative to parts being used in the newest iPhone (even though the iPhone clearly has much greater "demand").
Apple gets really good prices on their components even though they generate huge demand because the parts manufacturers know they'll be able to sell millions of whatever the part is, so they can afford to sell it with less margin for themselves per unit. This isn't true of smaller volume chips where the manufacturer has to price some amount of lost opportunity cost into a relatively small market item.
Low end micrcontrollers optimize low-power, manufacturing process support for analog blocks and flash memory and low cost per each square milimeter of silicon for low transistr count designs. You can best achieve all these using old manufacturing processes so moore's law really doesn't apply.
Following Moore's law tends to optimize $/transistor and low power consumption for high transistor count devices.
Chip prices are a function of supply v. demand and the supplier's learning curve as a function of cumulative volume (ie, yields). Chip prices DO increase as they approach the natural end of their lives due to other suppliers dropping out.
I really like Octopart and this feature is great. But was I the only one expecting to find out how much a 6502 cost in 1975? Or how much the original x86 chips cost on launch? Was the 80386 cheaper than the 8086 when they were introduced?
Seems like average price is the wrong metric. You're vulnerable to errors in your database like vendors setting the price to $999 to indicate that it's out of stock or vendors never updating the price of products that don't move.
Lowest price from a place that lists it as in stock seems like a far better metric. Or even something like 25th percentile price.
We've actually tried a bunch of different algorithms for pricing and we're still figuring out the best formula.
It's very atypical for a vendor to set the price to $999 to indicate OoS, since they do provide us with quantity information and can also choose to delist parts from their data feed. We generally get pretty good data from vendors and we have internal consistency checks to filter out the few odd cases where something doesn't line up.
