Taken together, the period's news has one subject, and it isn't language models: it's infrastructure. A billion-scale national plan, a European commitment to sovereign capacity, public funding for interconnect, and strategy reorganised by compute scarcity.
The shift in focus describes market maturity. While the gap between models was large, the contest was over model capability; with several competent models available, the contest became about who can run them at scale and at what cost.
Other sectors have lived through that transition. When the product becomes an ordinary commodity, value migrates to logistics, distribution and cost of operation.
For anyone building products, the practical consequence is that model choice lost relative weight against the choice of where and how to run it, with what fallback and at what cost per task.
And the question organising any operation stays the same, with a harder answer: when the preferred supplier doesn't deliver, is there an alternative already tested?