The reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist and to obtain profit that is mathematically impossible.
The reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been…
This is the cleanest summary of the AI hardware bubble. We’re pre-paying for a future supply chain, future workloads, and future profits that may never converge. It’s not RAM scarcity it’s speculative coordination failure at industrial scale.
Speculation is fine until the entire supply chain starts optimizing for a world that only exists on pitch decks. At some point, physics, cash flow, and grid capacity pull everything back to reality.
Being a non-native English speaker and having monaural hearing¹, the first time visiting the USA I thought they mentioned Empirical Units² when they tried to explain miles, feet and other measurement units they use on their island.
Then I learned they are in fact United States customary units but in the USA, they actually name those Imperial Units, implying that the UK still has a very strong influence on the USA. In reality, there are differences³ between Imperial Units and United States customary units to keep things in the USA practical (or lazy if you want), so I will keep calling their system Empirical Units as it is more fit for purpose (can’t name them Freedom Units any more given their Project 2025 regime).
Anyway, quite a while ago there was this cool XKCD “The Maritime Approximation” (image on the right) including only Imperial Units holding for Empirical Units as well: π mph ≈ 𝑒 kn (let’s use ISO unit symbols here, shall we) which is correct to < 0.5%.
Recently, I learned that with the same accuracy, there is a golden ratio between metric and Imperial/Empirical Units: 𝜙 km = 1 mi, also correct to < 0.5%.
Kevlin Henney wrote two great blog posts on these explaining way more background information:
I feel old, because I vividly remember the PCX (1985) graphics file format: it was the defacto standard under DOS.
TIFF (1986) was slightly younger, and came from the scanner background resulting in very large files though unlike PCX (which had lossless compression), TIFF supported both lossless and lossy compression.
On Windows and OS/2, you had BMP (1985, lossless initially only black and white).
All three suffered from the same problems: different implementations causing all sorts of compatibility problems
Those were the reason for the implementation of newer file formats for graphics like JPG (1992, lossy) and PNG (1996, lossless).
Often connections are TCP based, but sometimes UDP is all you have to test with, so I was quite surprised that testing that was quite forward. The solutions by [Wayback/Archive] How to Do a UDP Ping in Linux works on any platform where you can have nmap or netcat on installed (which by now is almost all platforms including Windows):
Everytime when installing a pfSenserouter from scratch, I seem to re-learn a few of the below quirks. So it was finally time to document them (:
Quite a few of my pfSense configurations are just doing routing between various networks, should not provide DHCP leases and do not always need or have a WAN connected (i.e. they are LAN-only).
Being a non-native English speaker and having monaural hearing¹, the first time visiting the USA I thought they mentioned Empirical Units² when they tried to explain miles, feet and other measurement units they use on their island.
The Wiert Corner - irregular stream of stuff 