Five signs that Generative AI is losing traction (garymarcus.substack.com)

An opinion piece argues that generative AI is showing early signs of losing traction: usage metrics and downloads are slowing, public product integrations are being questioned, and the economic logic for massive infrastructure bets may be shakier than assumed. Cited evidence includes declining mobile-download growth and daily use for ChatGPT (Apptopia/TechCrunch), a Census-based trendline highlighted by Apollo Global’s Torsten Slok, Airbnb’s CEO saying ChatGPT isn’t “quite robust enough” for travel integration, and anecdotal examples like dwindling interest in “vibe coding” and speculative social apps (the piece predicts an Sora 2–driven TikTok rival would see a similar boom-and-bust). For the AI/ML community this matters because it flags two technical and economic inflection points: first, simple extrapolations of LLM “scaling laws” may be encountering diminishing returns—bigger models and more compute no longer guarantee proportional user value; second, product-level robustness, UX, and real-world task fit are becoming the gating factors for adoption. The piece warns investors and builders against committing to endless capital-intensive infrastructure based solely on historical growth curves, arguing that irrationally large capex could turn into massive capital destruction if user engagement and model utility plateau.