AI Pulse July 28, 2026

EXECUTIVE SUMMARY

• Coverage of the technology itself has been recast as a story about physical inputs. Perscient's semantic signature measuring the density of language asserting that memory-chip shortages are slowing AI growth is the densest frame in the entire set and grew denser again this week, joined by rising density in language attributing slowdowns to data center construction delays, grid interconnection queues, and even a re-emergent GPU-scarcity frame. Reporting on memory pricing, gigawatt-scale load-shedding events, and multi-year interconnection timelines has displaced discussion of what models can do with discussion of whether the inputs exist to run them.
• The multi-decade payoff argument thinned out rather than being argued down. The signature tracking predictions of an AI-driven long-term investment supercycle recorded the sharpest weekly decline in the file, while language asserting that businesses increasingly doubt large AI spending grew markedly denser and language claiming that AI advances are translating into company profits fell. Alphabet's capex raise, negative free cash flow, and share-price punishment supplied the template for a press narrative in which capex hikes are no longer read as proof of demand.
• Scale conviction and payoff conviction are now moving independently, and the skeptical case is being made in capital-allocation terms rather than hype-cycle terms. Language asserting that infrastructure spending is massive and increasing held firm, supported by trillion-dollar sizing coverage, even while frames comparing AI capex to 1990s telecom overbuilding, alleging failed corporate initiatives, and flagging concentrated market valuation all eased. Commentary focused instead on purchase commitments, optionality, and cash-flow discipline, and balance-sheet storylines crowded out the labor-market storylines whose signatures declined together.
• Leadership language has decoupled from usage share. The signature tracking claims that Anthropic or Claude leads the competition remains far ahead of every rival frame and grew denser still on enterprise API share, the Claude Opus 5 release, and the expanded Cognizant partnership, even though OpenAI retains a consumer reach advantage of roughly twenty-five to one. Language claiming that the eventual winner has not yet been founded stayed flat, so the contest is being narrated exclusively among named incumbents.
• The Chinese challenge returned to the conversation on cost rather than capability, tying the competition frame back to both constraint and capital narratives. The signature tracking assertions that DeepSeek or China leads posted the largest weekly increase in the file on coverage of GLM-5.2, Kimi K3, and LongCat-2.0, framed around models that make comparable work an order of magnitude cheaper. The frame linking national energy-building capacity to AI leadership stayed elevated, connecting the rivalry directly to the power constraints of Section 1, while application-layer narratives covering medicine, education, and search remained well below their long-run levels.