American Science Policy and AI-Driven Research Reform

Summary

On August 4, 2026, Moonshots with Peter Diamandis featured Michael Kratsios, director of the White House Office of Science and Technology Policy (OSTP). Kratsios argues that declining U.S. scientific productivity stems from outdated grant structures and risk-averse review processes, and he outlines reforms including long-duration grants, AI-native laboratories, and open federal research data. The conversation offers a rare inside look at how the current administration is attempting to restructure science funding and technology policy amid intensifying competition with China.

Take-Home Messages

1. Funding Reform: Long-duration grants and rapid proof-of-concept reviews aim to realign federal science funding with the actual pace of discovery.
2. Risk Tolerance: Discretionary "golden ticket" funding mechanisms are designed to counteract consensus-driven review's bias against unconventional research proposals.
3. AI-Native Labs: Autonomous, continuously operating laboratories could compress experimental cycles, though they remain constrained by available hardware.
4. Data as Infrastructure: Decades of unstructured federal research data represent an untapped asset for accelerating AI-assisted scientific discovery.
5. Global Competition: Export controls, open-source AI ecosystems, and robotics supply chains are converging into a single strategic competition with China.

Overview

Michael Kratsios, director of the White House OSTP, joins Moonshots with Peter Diamandis to discuss why U.S. scientific productivity has stagnated even as research budgets have grown. Kratsios argues that the primary obstacle is not funding but the rigid, decades-old structure of how the government administers grants, reviews proposals, and manages scientific careers. His remarks draw on the White House's recently released Golden Age Report and the Genesis Mission, both of which he helped author.

Kratsios cites a National Academies finding that researchers spend nearly half of grant-funded time on administrative paperwork rather than experimentation, and he notes that a senior NIH scientist's median age is 71 despite Nobel-winning work typically occurring in a researcher's twenties. He attributes these patterns to standardized eighteen-month grant cycles and consensus-based peer review, both of which he says favor safe, incremental proposals over ambitious ideas. In response, the administration has begun piloting five-year grants funded in full at the outset, fast-track reviews modeled on pandemic-era funding, and "golden ticket" mechanisms that let individual reviewers unilaterally fund unconventional proposals.

Kratsios acknowledges that government institutions change slowly and expresses some skepticism about predictions of tenfold productivity gains, even as he endorses raising the Genesis Mission's stated goal from doubling to a tenfold increase in scientific output. He also concedes that autonomous, AI-run laboratories remain constrained by the physical limits of existing hardware rather than by AI capability itself, and that expansion will likely begin in narrow experimental domains before generalizing. On education and public trust, he distinguishes between technologies the government should leave alone and those requiring active intervention, arguing that AI's negative public image stems from a narrative shaped by early dystopian framing rather than the technology's demonstrated effects.

The conversation closes on the geopolitical stakes of AI leadership, where Kratsios frames semiconductor export controls, open-source model development, and humanoid robotics manufacturing as a single strategic competition with China. He credits 2019 restrictions on EUV lithography exports with slowing China's chip development, while acknowledging that restricting Nvidia's chip exports more recently carries the offsetting risk of accelerating Chinese self-sufficiency. Kratsios ties this competition to domestic policy choices, including regulatory sandboxes that let states and cities attract innovation-friendly investment, suggesting that America's science funding reforms and its international technology strategy are, in his view, part of the same effort to sustain leadership.

Implications and Future Outlook

If federal agencies scale the funding reforms discussed in this episode, researchers could see meaningfully more flexibility in how and when they receive support, though success will depend on evidence from the newly launched meta-science pilots rather than on the reforms' underlying appeal. Universities and national laboratories will need to make early decisions about investing in autonomous laboratory hardware, a costly bet given that the technology's near-term scope remains limited to narrower experimental domains. Agencies will also need better data on how AI is affecting the labor market, since current information is described as insufficient to guide targeted retraining or assistance programs.

Broader adoption of AI-ready federal research data will require governance decisions about who captures the financial benefit when public data enables private breakthroughs, a question the episode raises but does not resolve. Export control policy will likely remain a central and contested tool of technology strategy, with future decisions needing to weigh the demonstrated success of past restrictions against the risk that overly broad controls accelerate the self-sufficiency of competitors. Domestic manufacturing capacity in robotics and advanced chips will take years to build regardless of policy support, meaning near-term supply chain gaps with China are likely to persist.

Public trust in AI will remain a constraint on policy ambition unless government and industry succeed in shifting the dominant narrative away from job loss and existential risk toward tangible benefits in healthcare and other visible domains. State-level regulatory competition is likely to intensify as more jurisdictions seek to attract AI, robotics, and data center investment, which could accelerate innovation in permissive regions while raising oversight questions elsewhere. Over the coming years, the extent to which the administration's science and technology missions succeed will hinge less on total funding than on whether these more experimental funding structures can be sustained through changes in political leadership.

Some Key Information Gaps

1. What structural funding reforms most effectively increase discoveries per research dollar? Systemic significance across nearly all national research funding systems makes this question directly relevant to institutional investment strategy for years to come.
2. What governance frameworks are needed before scientific data becomes broadly AI-accessible? Cross-jurisdictional data governance choices will determine research pace and public trust across every field that depends on federally generated data.
3. Under what conditions do export controls accelerate rather than suppress a rival's technological development? This question carries direct trade, security, and industrial policy consequences that extend well beyond any single technology sector.
4. Who should bear the financial benefit when publicly funded data drives commercial breakthroughs? The answer shapes incentive structures for public investment across biomedical, energy, and materials science research alike.
5. What safeguards prevent a race to the bottom when jurisdictions compete on regulatory laxity? Jurisdictional regulatory competition is a generalizable governance problem relevant to energy, transportation, and data infrastructure policy alike.

Broader Implications

Institutional Path Dependency in Research Governance

Federal science funding institutions built around mid-twentieth-century assumptions about grant cycles, peer review, and researcher career progression tend to exhibit strong path dependency that resists incremental reform. Meta-science experimentation represents an attempt to introduce feedback loops that could gradually shift institutional equilibria toward evidence-based funding design, though such shifts have historically taken many years to consolidate. Whether these pilot programs scale into durable institutional norms will likely depend on whether early results generate enough political and administrative momentum to survive changes in leadership.

Principal-Agent Tensions in Data Stewardship

Treating federally generated scientific data as a public good creates a principal-agent problem between the taxpayers who funded its production and the private actors best positioned to capture commercial value from AI-driven breakthroughs built upon it. Absent clear stewardship rules, downstream benefits could concentrate among firms with the AI infrastructure to exploit newly released data rather than the public that financed its creation. How governments resolve this asymmetry over the coming years may determine whether public science functions as shared infrastructure or as a subsidized input to private capital formation.

Regulatory Arbitrage and Sovereign Competition

Jurisdictional competition over autonomous vehicle, drone, and data center regulation mirrors familiar regulatory arbitrage dynamics in which capital and firms migrate toward the most permissive rule sets. This dynamic plausibly extends beyond individual states into a broader contest between nations over which regulatory regime can attract frontier research and infrastructure investment. Sustained arbitrage could eventually pressure more cautious jurisdictions to loosen oversight or risk losing investment, gradually reshaping the balance between innovation speed and public accountability.

Export Controls as Reflexive Industrial Policy

Restricting semiconductor exports to a strategic rival functions less as a static barrier than as a reflexive intervention that reshapes the rival's own industrial incentives, and it may accelerate the very domestic capability the restriction sought to contain. This pattern reflects a broader dynamic in which unilateral technology restrictions can generate second-order capital reallocation toward substitute development within the targeted economy. Policymakers will likely need to keep weighing near-term capability suppression against the longer-run risk of catalyzing durable, self-sufficient competitors abroad.