Public Risk, Private Reward

In the AI Story

The pharmaceutical industry provides the clearest precedent. The mRNA vaccine technology commercialized by Pfizer and Moderna during COVID-19 rested on three decades of publicly funded research, largely at the University of Pennsylvania, where Katalin Karikó conducted work the private sector considered so unpromising she was repeatedly denied tenure. Conservative estimates place total public investment in mRNA-related research at over thirty billion dollars. Moderna and Pfizer's combined COVID vaccine revenue exceeded seventy billion dollars. The NIH received no direct equity return, no royalty stream, no financial participation in the commercial success.

The AI industry follows the same template. Two AI winters — 1974–1980 and 1987–1993 — saw private funding collapse while public institutions sustained the foundational research. Geoffrey Hinton's deep learning work at Toronto, Yann LeCun's convolutional networks at Bell Labs and NYU, Yoshua Bengio's attention mechanisms at Montreal — all substantially funded by NSF, DARPA, and their Canadian and European equivalents through decades when private capital considered the work commercially worthless.

The private sector entered in force only after the risk had been substantially reduced. Google's acquisition of DeepMind in 2014, Facebook's hiring of LeCun that same year — these occurred after publicly funded research had demonstrated technical viability. The salaries became private. The intellectual capital was substantially public. The Bayh-Dole Act of 1980 was designed to solve a commercialization problem, not a distribution problem. Patents are licensed for fees that represent a fraction of commercial returns. The public bears the full risk of foundational research and receives a minimal share of financial upside.

The AI transition makes this pattern more consequential than any previous technology cycle because the scale of returns is larger (combined AI market capitalization exceeds pharmaceutical or semiconductor industries), the speed of transition is faster (years rather than decades to commercialize publicly funded deep learning), and the concentration of returns is more extreme (winner-take-most dynamics amplified by network effects).

Origin

The pattern has been documented by innovation historians for decades — Vannevar Bush's 1945 Science: The Endless Frontier anticipated the distributional question that Mazzucato's framework would later diagnose. David Noble's archival work on military funding of industrial R&D, Fred Block's studies of the developmental state, and the Janeway framework for public-private innovation all contributed empirical foundations.

Mazzucato's contribution was synthesis and prosecutorial specificity: tracing the public funding behind specific commercial products with enough granularity to make the general pattern unmistakable, and developing the analytical framework for what the institutional response should be.

Key Ideas

Asymmetric time horizons. State operates on decades; private capital on quarters. The foundational research falls in the gap.

No return mechanism by design. The institutional architecture contains no pipe for returns to flow back to the public investor.

mRNA precedent. Thirty billion in public investment, seventy billion in private returns, zero direct public participation.

AI amplification. Scale, speed, and concentration of returns make the pattern more consequential in AI than in any previous cycle.

Bayh-Dole inadequacy. The 1980 legal framework for public-funded patents solved commercialization but not distribution.

Debates & Critiques

The debate centers on whether mechanisms for public return would reduce private incentives to commercialize publicly funded research. Empirical evidence from Finland's Sitra, Israel's Yozma, and Singapore's Temasek contradicts the claim that public return mechanisms suppress innovation.