American Science Doesn't Lack Money, It Spends It Wrong
The NIH budget went from $14 billion in 1998 to $47 billion in 2024, yet output efficiency halves every 9 years. Kratsios's answer: the problem isn't the budget number, it's the funding mechanism itself.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The NIH budget tripled while output efficiency halved every nine years
Kratsios cites Eroom's Law: the NIH budget was $14 billion in 1998, $27 billion in 2003, $47 billion in 2024, but breakthrough therapies have not grown proportionally, and output efficiency per dollar has fallen roughly 80-fold since 1950 and halves again every 9 years. His explanation is not that "the easy problems have all been solved," but that the funding mechanism has never been innovated — "the answer has always been to keep doing the same thing, just with more money, and then hope output increases proportionally." His example is flight speed: in the 1990s you could fly Concorde, and now it's actually slower.
— Michael KratsiosGolden tickets: let a single reviewer decide unilaterally
NSF's current peer review has three or four reviewers in the same field collectively pick the projects "most worthy of funding," with the result that scientists only propose safe ideas that land in the "strike zone." Kratsios says he wants to pilot golden tickets at NSF: each reviewer gets one to three tickets and can unilaterally select a project without committee approval. The mechanism was first tested in places like Denmark. A side benefit is that it attracts better reviewers to participate — because they now hold this ticket. At the same time, he wants to experiment with grants of different durations: NSF grants are now almost all 18 months, but some ideas can be validated in 3 months or 6 months, and some need 5 years.
— Michael KratsiosFund people rather than projects
Kratsios endorses the venture capital logic: find the right person, give them far more money than they applied for and a very long time horizon, and don't watch how every dollar is spent. He cites NSF's GRFP fellowship as an example — each year it goes to about 2,600 of the best prospective PhD students, the money is fully portable, students can take it to any school, and universities compete for them. Another path is big projects: the Manhattan Project, Apollo, the Human Genome Project. He acknowledges that China is very good at big projects under a central planning system, but argues that the U.S. needs to restore this capability, not do only this one thing.
— Michael KratsiosA quantum computer by 2028, fusion by 2035
Kratsios lists this administration's big bets: put Americans back on the moon by 2028, put a nuclear reactor in space, and build the first components of a lunar base by 2030; last month the president signed an executive order launching a new national quantum initiative, with the goal of building a "scientifically relevant" quantum computer by 2028, directed to the Department of Energy; the fusion target is 2035. The flagship program is called the Genesis Mission, whose core is to use AI to double America's scientific output. He states explicitly that the government will not compete with private companies on AI compute, but will get scientists doing basic research to think about how to use AI to accelerate their own work.
— Michael KratsiosInstitutional dispersion is a feature, not a bug
Some argue for merging science agencies into a single department; Kratsios is opposed. His evidence is China: since the U.S. imposed export controls in 2019, China has been working to crack EUV lithography and has not broken through — "nothing is more important to their economy than solving this scientific problem, and they haven't been able to do it." He argues that the competition among agencies in the U.S. system and its free-market-style innovation paths are "a feature rather than a bug," and that having DARPA people sit inside the Defense Department rather than in some science agency actually makes their work fit their mission better.
— Michael KratsiosVannevar Bush's model is seventy years out of date
In 1945 Vannevar Bush's reply to Roosevelt, Science: The Endless Frontier, established the current system: government pays, universities do basic research. Kratsios points out that at the time about 70% of R&D was funded by the federal government and 30% by private companies, and seventy years later it has completely reversed — now about 70% of R&D is done by private companies and 30% is funded by the federal government. At the same time, philanthropically funded focused research organizations like Howard Hughes and Max Planck have emerged. His conclusion is that Bush's model "isn't as applicable today as it was then, and needs refreshing." The new first principle is: the government should only pay for what private companies and philanthropy have no incentive to do.
— Michael KratsiosSeven in ten STEM PhDs are not American
Kratsios says this is a number he has tracked for years: in computer science, thirty years ago about 70% of PhDs were American and 30% foreign, and now it has completely reversed. He calls the figure that "seven in ten STEM PhD candidates are not American" something that should alarm everyone. He mentions that NSF once had programs funding gifted American high school and middle school students, which were later shut down, and he sees this as part of the DEI tide — San Francisco canceling advanced algebra in high school math and the University of California system's fight over the SAT are the same kind of phenomenon. He argues for rewarding America's most talented students again.
— Michael KratsiosFauci has done more damage to science than anyone in modern times
Kratsios traces the collapse of trust in science to the COVID period: advocating that students wear masks in school while the pediatric association sent a letter saying they could protest in groups but not visit a dying grandmother in the hospital — "any median ordinary American hearing this can no longer take science seriously." He believes Fauci "has done more damage to science than anyone in modern history," and says it will take a long time to repair. The solution he repeatedly stresses is a return to "gold-standard science": science is a process of constant experimentation and learning in which everyone participates, not dogma handed down from on high.
— Michael KratsiosIn their own words · checked verbatim
the answer has always been let's just keep doing the same thing but add more money and hope that we get more outcomes proportion rather than asking the harder questions about are there other ways that we could be conducting the science
Michael Kratsios15:08
they have been trying to essentially figure out how to do EUV lithography since since we put the export controls in 2019. No breakthrough has happened. There's nothing more important to their economy than solving that scientific problem and haven't been able to do that.
Michael Kratsios27:12
Now roughly 70% of R&D is done by by the private sector and 30 is funded by the federal government.
Michael Kratsios31:13
the idea that, you know, seven out of 10 STEM related PhD candidates in the United States are not American, like anyone can look at that and say like that just doesn't doesn't look right
Michael Kratsios41:23
he did more disservice to science than than anyone in in modern history.
Michael Kratsios45:27
it's not some sort of like dogma or rule or some edict that comes from on high but it's a a process that all of us are part of that we're all experimenting and learning about the world and the universe that we live in.
Michael Kratsios46:28
the median age of an intramural researcher at NIH is 71 years old.
Michael Kratsios49:30
Figures
| Eroom's Law efficiency decay | Output efficiency per dollar has fallen about 80-fold since 1950, halving every 9 years | 14:08 |
| NSF annual budget | about $8-9 billion | 7:06 |
| Number of GRFP fellowships | about 2,600 per year | 21:11 |
| Change in federal government share of R&D funding | from about 70% to about 30% | 30:13 |
| Median age of NIH intramural researchers | 71 | 49:30 |
| Global population cooking with polluting fuels | about 2 billion people, with about 3 million deaths per year | 11:06 |
Glossary
- Eroom's Law
- Drug development efficiency halves every 9 years, the opposite direction from Moore's Law.
- golden tickets
- Giving each reviewer a few tickets that can unilaterally approve funding, bypassing committee consensus.
- RCP 8.5
- The highest-emissions extreme scenario in climate models, recently withdrawn by the IPCC.
- GRFP
- NSF's flagship PhD fellowship; the money is portable and students can choose their own school.
- Genesis Mission
- The flagship program to accelerate scientific discovery with AI, aiming to double U.S. scientific output.
- OSTP
- Established in the 1960s, it coordinates science and technology policy across the federal government and sets the national science and technology agenda.
How to listen
Founders and investors watching U.S. research funding mechanisms and the direction of science and tech policy; anyone who wants to understand the policy-side logic of U.S.-China tech competition.
8:16-13:08 the climate science section — conclusion first, limited information gain.