Generative AI is supposed to help anyone innovate — but is it mainly the big, research-heavy companies that actually profit?

This explores whether generative AI's promise to put innovation within everyone's reach matches who is actually capturing its value right now, with the suspicion that large, R&D-heavy firms are pulling ahead. One caveat up front: the corpus has no direct firm-level data on that concentration. It does, however, explain why a technology can spread widely and still pay off narrowly.

The democratization case is real, and it is strongest at the level of individuals. Does generative AI enable exploratory innovation beyond predictive optimization? argues that GenAI breaks from earlier predictive AI because its chat interfaces and low cost let almost any organization experiment, not just firms with data science teams. Adoption numbers back up the access part of that claim. Is generative AI adoption now outpacing the personal computer? finds 54.6% of Americans using GenAI three years after ChatGPT launched, well ahead of the PC (19.7%) and the internet (30.1%) at the same point. Workplace use has also moved past where the PC was at the same age.

Wide use is not the same as wide gains, though. The same researchers estimate the aggregate productivity gain at about 1.1% What is generative AI's actual aggregate productivity gain today?. Many people are using the tool, but so far it is adding little to total output. Does AI adoption require rethinking how organizations coordinate? offers one explanation. AI improves individual tasks, but it also makes decisions across an organization less consistent. The big returns only arrive when a firm reorganizes how those decisions fit together. That kind of reorganization takes management capacity, spare resources and tolerance for experiments, which are exactly what large R&D-intensive firms have. If that's right, concentration isn't a betrayal of democratization. It's what you'd expect when access is cheap but turning access into value is expensive.

Two more threads suggest the gap could widen. Does automating labor concentrate power more than augmenting it? warns that AI built to replace workers rather than help them shifts bargaining power to whoever controls the technology, and that this can lock into a stable trap. On the individual side, Does passive GenAI use risk causing worker deskilling? shows that using GenAI passively can erode skills. The same tool that lowers the barrier to entry can quietly hollow out the expertise that smaller players would need to compete.

The most useful reframe comes from Does generative AI inevitably worsen or reduce inequality?. Across work, education, health and information, GenAI has both widened and narrowed inequality, and the deciding factors were access, how well it was integrated, and incentives, not the capability itself. So the democratization claim isn't simply true or false. It describes what the technology makes possible, while concentration describes how it is currently being deployed. The real question is whether smaller firms can get the complementary assets: the organizational redesign, the skills and the bargaining power. The model alone won't decide that.