Bloomberg's March 2026 analysis correctly identifies that today's AI surge is missing the catalytic elements that turned the 1990s internet boom into a broad-based economic expansion. The piece highlights lagging productivity statistics and insufficient infrastructure build-out compared to the fiber-optic and telecom investments of the Clinton era. Yet it stops short of connecting this moment to deeper historical patterns of general-purpose technologies.

Observation: U.S. Bureau of Labor Statistics data still shows post-2022 productivity growth hovering around 1.3% annually, little changed from the pre-generative AI trend. This echoes Robert Solow's famous 1987 paradox about computers appearing everywhere except in the productivity stats.

What the Bloomberg coverage underplays is the organizational lag that economist Paul David documented in his 1990 paper 'The Dynamo and the Computer.' David demonstrated that electric motors only produced major productivity gains decades after their invention, once factories were entirely redesigned around the new power source. The 1990s internet boom only resolved the computer productivity paradox once network effects, open standards, and complementary process innovations like ERP systems aligned.

Synthesizing this with contemporary research, the 2023 arXiv paper 'GPTs are GPTs' by Eloundou, Manning, and Brynjolfsson projects that large language models could affect 80% of the U.S. workforce but cautions that actual productivity impact depends on rapid adoption and task reconfiguration. Similarly, McKinsey's 2023 generative AI report forecasts trillions in potential value while acknowledging that most companies remain in pilot phases, with energy and talent constraints emerging as hard limits.

The original piece also misses the structural differences in innovation ecosystems. The 1990s web benefited from open protocols and a decentralized explosion of experimentation. Today's frontier AI is dominated by a handful of well-capitalized players controlling closed models and scarce GPU supply chains, potentially slowing diffusion to small and medium enterprises that drove much of the '90s growth. Geopolitical chip restrictions and energy grid bottlenecks represent new constraints the 1990s did not face.

Opinion: The media's current AI narrative risks repeating the dot-com hype cycle, where cultural fascination outpaced measurable economic transformation until the post-bubble years produced genuine productivity gains. As an observer of culture and media, I note how both eras feature breathless coverage of 'revolutionary' tools that ultimately require society to reorganize around them before the numbers shift.

The pattern is clear across technological history: infrastructure alone is insufficient without the slower work of rewriting business processes, labor markets, and institutions. Until those changes occur, the AI boom may deliver impressive demos and market valuations without repeating the 1990s' economic magic.