Taken together, the period's news describes one plot in different versions: the contest stopped being over who has the best model and became over who can compute.
On the supply side, fabs announced across three continents, a new country entering packaging, and switching platforms with double the previous capacity.
On the demand side, gigawatt-scale capacity contracts, custom-designed chips entering production, a billion-dollar financing guarantee for a data centre build, and a service selling spare compute.
In between, memory prices rising enough to appear as a separate line in the largest buyers' budgets, and manufacturer shares falling on doubt about how long that demand holds.
For smaller operations, the practical translation of those figures fits in one sentence: compute capacity became an input with a lead time, and inputs with lead times get contracted before you need them.