The market for computing capacity has grown quickly enough that its pricing conventions
have not caught up with its size. Buyers negotiate privately, sellers quote rates that few
customers actually pay, and the published figures that circulate in the press are usually
list prices rather than cleared transactions. This is a familiar pattern. Most commodity
markets pass through a period in which the headline number and the transacted number drift
apart, and the gap only closes once a benchmark emerges that participants trust enough to
reference in contracts.

What makes the current moment unusual is the speed of the transition. Physical commodity
markets took decades to standardize. Grain needed uniform grades before it could trade on
an exchange, and those grades required an inspection regime, a dispute process, and a set
of tolerances that everyone agreed to in advance. Crude oil needed a reference blend with
a specified density and sulfur content, because a barrel from one field is not
interchangeable with a barrel from another without an adjustment.

The lesson from those markets is that price discovery is downstream of unit definition.
An index that reports what buyers paid is only meaningful if the thing they bought is
consistent from one transaction to the next. Where the underlying good varies in quality,
the index inherits that variation as noise, and the noise is not random: it correlates
with whichever sellers happen to dominate the sample in a given period. A benchmark built
on transactions is a genuine improvement over one built on quotes, but it does not by
itself solve the problem of what, exactly, is being counted in the denominator.
