The individual
Portable fellowships and longer grants could give early-career researchers more control over their work.
A close reading of a White House science proposal
A golden agefor whom?
The White House calls the research system "calcified" and proposes new routes around universities, peer review, and public infrastructure. Those routes may cost the United States scientific capacity it cannot replace.
Read the analysisvChapter 01Diagnosis
The report diagnoses an American research system in decline.
Science: A New Golden Age argues for a new purpose, workforce, and system of judgment for American science.
Grant-writing burden and uneven access to research resources are real problems. The report adds slow translation, then argues that the postwar research system has exhausted itself.1
"We have become dependent on a narrow set of legacy institutions. Our incentive structures reward conformity over bold inquiry."
01
Our reading
The report may replace core institutions before alternatives prove their worth.
New funding models could widen access. Across its chapters, however, the report casts universities, peer review, broad portfolios, and public institutions as bottlenecks. Federal science relies on them to maintain facilities and train researchers. They also spread decisions across thousands of specialists.
IDiagnoseSlower progress
IIRewireFunding & institutions
IIIMobilizeIndustry & missions
IVDistributeSkills & regions
VAutomateAI-native science
We focus on downside risk and separate proposals from possible consequences. We label inferences as such. We checked every report excerpt against the supplied Markdown transcription and the original PDF.
Chapter 02Institution
Fault line 01
Individual fellowships still depend on institutions.
A grant cannot power a freezer or run a clinical trial on its own.
The invisible institution
Every grant relies on shared equipment, biosafety staff, secure computing, libraries, procurement, and people who keep complex facilities running.
The substitution problem
Federal dollars could follow individual researchers while universities receive less support for facilities and administration. Wealthy institutions can cover more of that gap than regional universities or emerging research institutions. Current financing therefore suggests that a policy intended to spread funding could concentrate it. This is an inference.
RESEARCHER
COREShared instruments
SAFESafety & compliance
DATASecure data
TEAMStudents & staff
OPSFacilities
XFERTechnology transfer
The proposal centers the researcher
48%
Higher education performs 48% of all U.S. basic research, the largest share of any sector.2
63%of academic R&D is basic research
55%of academic R&D is federally funded
26%is already financed by institutions themselves
"…rein in indirect cost recovery that supports administrative bloat, redirecting that money to real scientific infrastructure."
Inside indirect costs
A grant needs a building that works.
Federal rules classify shared research expenses that cannot be assigned to one project as indirect costs. Universities negotiate rates with the government, and auditors review them. Allowable costs include utilities, equipment depreciation, building operations, purchasing, and compliance.45
Administrative efficiency deserves scrutiny. Calling the entire category "bloat" hides the facilities and staff that federal projects use.
-80°CFreezersContinuous power
BSLBiosafetyContainment & training
IRBHuman researchEthics review
DATACybersecurityProtected records
COREInstrumentsShared maintenance
OPSFacilitiesAir, water & waste
Chapter 03Portfolio
Fault line 02
A "rebalanced" portfolio leaves a $28 billion question.
The report never defines what happens to applied research and development.
"Agencies should prioritize foundational research in the biological sciences over the life sciences, a broader category not focused on foundational research."
The annex calls for more physical science, computing, engineering, and early-stage biology. It gives no direction for translational work beyond that boundary.114
Federal agencies spent $45.6B on life-sciences R&D in FY 2024. Basic research received $17.6B, while applied research and development accounted for $28B. The latter includes drug safety, clinical research, population health, and implementation science.2
Federal life-sciences R&D / FY 2024$45.6B total
$17.6BBasic
$21.0BApplied
$7.0BDevelopment
$28B in applied R&D and development
This policy stress test applies the report's relative priority. The report assigns no dollar amounts.
Chapter 04Judgment
Fault line 03
Alternative peer-review experiments deserve a trial.
White House priorities could give fewer officials more control over funding.
The recommendation
The annex calls for government to "supplement conventional, consensus-driven peer review" with golden tickets, lotteries, prizes, regranting, and program-manager models. It also calls for controlled experiments to see which methods work.1
Verified excerpt / PDF p. 113 / print p. 95The risk
The report also asks program officers to set scientific direction and agencies to align investment with White House missions. Taken together, alternative review and centralized priorities could move decisions away from domain experts and toward a smaller number of political appointees and program managers.
Distributed review
1234567
v
Program officerexpert input + public criteria
NSF uses a program officer and 3-10 external experts to assess proposals against published merit criteria.8
SUPPLEMENT
Portfolio discretion
AIQuantumFusionSpace
v
Empowered managermission + portfolio + discretion
Program managers can make bold bets. Public criteria, conflict controls, and external review would limit the damage from one gatekeeper's blind spots.
Chapter 05People
Fault line 04
Research universities recruit the people America hopes to retain.
The country renews its scientific workforce through training and immigration.
The report calls for fellowships, early-career independence, and technical apprenticeships. U.S. universities also recruit, train, and retain a mobile global workforce.
In 2024, temporary visa holders earned 38% of U.S. science and engineering doctorates. Five years after graduating, 73% of earlier temporary-visa cohorts were still in the United States. Foreign-born scientists made up 46% of doctorate-level U.S. science and engineering workers.9
Cuts to university research, narrower fields, and unpredictable grants would give other countries an opening to recruit the same scientists.
38%S&E doctorates earned by temporary visa holders
73%still in the U.S. about five years later
46%doctorate-level S&E workers born abroad
THE U.S. TALENT PIPELINE
1RecruitGlobal students
2TrainLabs + mentors
3FundGrants + fellowships
4RetainCareers + visas
5Spill overFirms + regions
Chapter 06Machine
Fault lines 05 + 06
A market for scientific results could put private owners at every gate.
The report describes AI agents buying compute, data, and laboratory work.
"In such a world, experimental information becomes a tradeable commodity, and price mechanisms replace slow institutional coordination."
SCIENCE MARKET / CONTINUOUS
AGENT NETWORK 08:41:22OPEN BOUNTY$1,000,000
Validated therapeutic target / rare disease
3 autonomous labs bidding
ACCESS LAYER
Computemetered
Datalicensed
Roboticscontracted
Verificationrewarded
THE UPSIDE
Faster iteration and direct rewards for replication.
THE PUBLIC RISKAccess follows purchasing power; research agendas follow bounties; data and methods become toll roads.
600+research projects
A working public alternative
NAIRR shares compute with researchers nationwide.
NSF's National AI Research Resource gives researchers access to compute and models plus technical support. By 2026 it supported more than 600 projects and 6,000 students in all 50 states, D.C., and Puerto Rico.10
Public infrastructure can keep AI research open to scientists whose institutions cannot buy premium compute or data.
The unpriced input
AI runs on a physical system.
U.S. data centers used 4.4% of national electricity in 2023. DOE projected 6.7-12% by 2028.11 A credible national AI-science strategy must track model throughput alongside power use, water use, and regional grid stress.
4.4%20236.7%2028 low12%2028 high
051015%
ConclusionSafeguards
Six safeguards for reforms worth testing.
Test alternatives alongside existing review
Pilot new funding methods, compare results, and publish the evidence before agencies expand them.
Fund shared research costs
Publish facilities and administrative rates, then pay for the capacity federal projects use.
Keep the portfolio broad
Add physical-science capacity while protecting applied life science and public health.
Check program-manager power
Require public criteria, conflict rules, external review, and a way to appeal decisions.
Keep training and immigration stable
Support early-career grants, research visas, and regional institutions that anchor careers.
Build a public AI backbone
Give researchers shared compute and data, then account for energy use.
America's 1945-era research institutions need reform. Since then, universities and federal agencies have trained generations of scientists, spread research capacity across states, and turned federal investment into inventions that recipients chose to develop.12
Policies that strip capacity from universities, narrow the research portfolio, concentrate discretion, or put scientific exchange behind market gates may produce a few brilliant successes. The broader enterprise would have less capacity to train people, maintain facilities, or pursue work with no immediate buyer.
Documentation
Sources & notes
We accessed these data on July 23, 2026. Dollar figures are nominal unless noted. We use contemporary coverage for context and federal statistics for quantitative claims.
- 01White House, Science: A New Golden Age (2026)Primary document
- 02NCSES, State of U.S. Science & Engineering 2026Federal statistics
- 03NCSES, Higher Education R&D Survey, FY 2024Federal statistics
- 04Congressional Research Service, Universities and Indirect CostsCongressional analysis
- 05NSF, What are indirect costs?Federal guidance
- 06NIH, Budget and extramural research reachFederal data
- 07NIH, Direct economic contributionsFederal analysis
- 08NSF, Proposal processing and merit reviewFederal guidance
- 09NCSES, STEM talent and workforce, 2026Federal statistics
- 10NSF, National Artificial Intelligence Research ResourceFederal program data
- 11DOE / Berkeley Lab, U.S. data-center energy useFederal analysis
- 12
- 13Science, 'White House report calls for big changes...'Contemporary reporting
- 14The New York Times, 'Trump's Plan for Science'Contemporary reporting