The Structural Deficit of Big Tech's Artificial Intelligence

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The generative artificial intelligence systems deployed by major technology corporations between 2022 and 2026 rest on a foundation of structural contradictions that have matured from theoretical concerns into observable, quantified failures. This paper examines five interlocking crises: (1) the extractive and legally precarious data acquisition practices underpinning large language model training; (2) the provably irreducible architectural limits of the transformer paradigm; (3) the systematic collapse of enterprise deployment value creation; (4) the disintegration of any coherent path to profitability at the frontier; and (5) the emerging legal framework that threatens to retrospectively penalize and prospectively constrain the entire industry. Drawing on peer-reviewed research, institutional surveys, regulatory filings, and financial disclosures current through mid-2026, this analysis demonstrates that the dominant AI paradigm is not undergoing routine growing pains but a structural reckoning: the simultaneous exhaustion of its data substrate, the mathematical ceiling of its core architecture, the commercial failure of its enterprise proposition, the negative unit economics of its compute-intensive operations, and an escalating legal liability that is beginning to be priced in real settlements. The paper concludes that these are not independent risks but a single, compound structural failure demanding fundamental re-evaluation of both the technology and its economic premises.