Rebuilding America’s Public Research Capacity: Talent, Shared Infrastructure, and Open Evidence

Research capacity is a system, not a budget line

A country can increase research appropriations without increasing its ability to produce trustworthy knowledge. Funding may arrive after experienced staff leave. A grant can pay for an experiment while long-term data preservation remains unfunded. A new facility can be announced without the technicians, maintenance schedule or user program required to keep it productive. A government can commission advanced analysis and still lack enough internal expertise to evaluate the contractor’s result.

Public research capacity has at least five components: people, physical and digital infrastructure, durable evidence, institutional coordination and translation into use. Weakness in one can reduce the value of the others. More researchers cannot compensate for inaccessible equipment. More computing cannot repair poor data provenance. More publications do not guarantee that an agency can incorporate results into regulation, procurement or operations.

This paper focuses on the federal role because it supplies functions markets and individual universities do not reliably provide: support for basic research, continuity across commercial cycles, national facilities, mission-oriented work, common data assets, standards and the ability to coordinate knowledge for public purposes.

The United States is strong, but the structure of that strength matters

The National Science Board’s State of U.S. Science and Engineering 2026 describes a large and productive enterprise. U.S. R&D expenditures equaled an estimated 3.4 percent of gross domestic product in 2024. China and the United States together accounted for more than half of global R&D performance when adjusted for international comparability, with China estimated at $1.028 trillion and the United States at $1.009 trillion.

Those headline figures do not indicate that the systems are interchangeable. In the United States, business performed 77 percent and funded 75 percent of national R&D in 2024. Experimental development—work directed toward producing or improving products and processes—accounted for about two-thirds of total performance. This is consistent with a research system led in scale by firms pursuing closer-to-market applications.

The federal role is different. Federal agencies obligated approximately $194 billion for R&D in fiscal year 2024. Across fields, 24 percent went to basic research, 27 percent to applied research and 49 percent to experimental development. The federal government supplied 40 percent of domestic basic-research funding, more than any other sector, while business supplied 34 percent. Basic research represented only 6 percent of business R&D in 2023.

This division is not evidence that firms neglect science or that government should displace them. It shows comparative roles. Businesses are essential to development, scaling and commercial learning. Federal support is unusually important where benefits are diffuse, timelines are long, uncertainty is high or the public value cannot be captured by one investor.

Public capability begins with people who can exercise judgment

Scientific capacity is frequently discussed as the number of degree holders. Agencies also need people who can define a research question, manage a portfolio, maintain an instrument, assess a model, negotiate data rights, interpret uncertainty and determine whether a contractor’s evidence supports payment or deployment.

GAO has identified federal strategic human-capital management as a high-risk area since 2001. Its work on the science and technology workforce emphasizes three recurring needs: strategic planning to identify skill gaps, improvements to hiring and pay, and a work environment capable of retaining qualified people. The problem is not simply that government salaries differ from the private sector. Hiring timelines, rigid occupational classifications, uncertain project continuity and limited technical career paths can make federal service difficult to enter or sustain.

The federal government should not attempt to match every private compensation package. It can compete through mission, access to unique facilities and data, professional autonomy, public impact and career stability. Those advantages require credible management. A scientist recruited for difficult mission work will not stay if procurement delays prevent tools from arriving, leadership repeatedly changes priorities without explanation or advancement requires abandoning technical practice for general management.

A government-wide science and technology skills map

Agencies should maintain a shared, annually updated map of mission-critical technical capabilities. The unit of analysis should be a capability, not merely an occupational series. “Artificial intelligence” is too broad; an agency may specifically need model evaluation, secure data engineering, accelerator procurement or causal inference. “Biology” may conceal distinct needs in field surveillance, genomic methods, laboratory safety or regulatory science.

The map should identify current staffing, retirement exposure, hiring time, contractor dependence, geographic concentration and the missions placed at risk by a gap. Aggregated results should be public where security permits. Agencies could then share talent pipelines, training and temporary assignments instead of competing separately for every scarce skill.

Workforce planning must include technical support roles. A national laboratory cannot operate advanced equipment without technicians and skilled trades. A data program cannot remain reproducible without librarians, archivists and data engineers. The 2026 indicators report that STEM middle-skill occupations declined as a share of the total workforce between 2000 and 2024 even as science and engineering occupations grew. Capacity policy that focuses only on advanced degrees will miss the people who keep facilities operational.

Use multiple pathways into and through public service

No single hiring authority can solve the workforce problem. Agencies should use a portfolio:

• Direct-hire authority for documented mission-critical gaps, with public reporting on time to hire and retention.
• Term appointments and technical fellowships for emerging fields, paired with clear ethics and post-service rules.
• Rotations among agencies, national laboratories, universities and state institutions that preserve employment rights and security requirements.
• Paid pathways for technicians, community-college graduates and workers moving from adjacent industries.
• Senior technical career tracks that allow advancement without requiring a switch to general administration.
• Shared expert pools for evaluation, procurement and incident response that smaller agencies could not sustain alone.

Each pathway should be evaluated for conversion, retention, performance and distribution across agencies—not only the number of people admitted. A fellowship that trains talent for rapid departure may still create public value, but its purpose should be explicit.

Treat shared research infrastructure as a public utility for discovery

Many research tools are too expensive or specialized for every institution to own. Shared user facilities, repositories, testbeds, telescopes, research vessels, advanced computing and secure data environments expand the number of researchers able to attempt consequential work.

The National Artificial Intelligence Research Resource offers a current example. Launched as a pilot in 2024 and led by the National Science Foundation, NAIRR connects researchers and educators with computing, data, models, software, training and expertise. NSF reports that the initiative has supported more than 600 research projects and 6,000 students across all states, the District of Columbia and Puerto Rico, with participation from 13 federal agencies and 28 nongovernmental partners.

The relevant lesson is not limited to AI. Shared infrastructure works when access, support and governance are designed together. Providing compute credits without expert help may advantage established teams that already know how to use them. Providing a repository without durable metadata can create a large but unusable collection. Public-private contributions can expand capacity, but terms must address research independence, security, publication and continuity if a contributor withdraws.

Congress and agencies should require lifecycle plans for major research infrastructure. A plan should cover capital cost, staffing, maintenance, cybersecurity, user access, data stewardship, decommissioning and the consequences of an interruption. New construction should be compared with upgrades, shared access and networked facilities rather than evaluated as a stand-alone symbol of commitment.

Fund the evidence after the experiment ends

Research data are not self-preserving. Files require documentation, durable identifiers, formats, privacy review, security controls, repositories and people responsible for maintenance. Software and computational environments change. A result that cannot be reconstructed may remain interesting, but it is less useful for cumulative science or public accountability.

Federal grant and contract budgets should treat data and software stewardship as eligible research costs from the beginning. Data-management plans should identify which outputs will be public, which require controlled access, which cannot be shared, how long they will be retained and who pays for continued access. Security, privacy, Indigenous data governance, confidential business information and export controls can justify restrictions; “open” should not mean indiscriminate release.

The 2022 Office of Science and Technology Policy public-access memorandum directed agencies to update policies so federally funded research articles and supporting data become publicly accessible without an embargo, subject to lawful limitations. Implementation should be judged by usability as well as formal availability. A PDF and an undocumented data file technically satisfy access while leaving important results difficult to verify or reuse.

Protect continuity without protecting every program forever

Research requires the possibility of failure. It also requires enough continuity to distinguish failure from interruption. Abrupt funding gaps can destroy samples, disperse teams, terminate longitudinal data and waste prior investment. At the same time, permanent continuation without review can preserve low-value programs.

Major portfolios should use staged, multi-year commitments with defined review points. Agencies should report the costs of starting, pausing, restarting and terminating work. Review should consider scientific quality, mission relevance, portfolio balance, replication, infrastructure dependence and opportunity cost. When termination is appropriate, wind-down funds should preserve records, samples, software and transition obligations.

Coordination is especially important because six federal departments and agencies accounted for 95 percent of federal R&D obligations in 2024. Concentration can produce scale and expertise, but it also means that a decision in one organization can reshape a field or facility network. Cross-agency road maps should reveal dependencies without imposing one centralized scientific agenda.

Measure translation as capability and use

Publication and patent counts are informative but incomplete. A public research portfolio can also create standards, validated methods, trained personnel, shared datasets, operational models, clinical or regulatory evidence and the ability to respond faster during a crisis. These outputs may not generate an immediate licensing event.

Programs should select measures that match their theory of public value. Examples include external use of a dataset, replication of a finding, time saved in a public process, adoption of a standard, users served by a facility, movement of trained personnel into mission roles, documented policy changes and options preserved for future research. Metrics should include negative results and stopped projects when they prevent repetition of an unproductive path.

Not every outcome can be attributed cleanly to one grant. Evaluation should combine quantitative indicators with independent portfolio review and should avoid rewarding easily counted but low-value outputs.

The WSOI public research capacity agenda

Research administration is scientific capacity

Researchers experience public policy through grant systems, contracting, security review, information technology and access agreements. Administrative controls protect important interests, but fragmented or unpredictable processes consume scientific time and can exclude smaller institutions that lack specialized compliance teams.

Agencies should measure the time from a complete proposal to decision, award to usable funds, equipment request to delivery and approved data request to access. Delays should be separated by cause so that safeguards are not blamed for ordinary process failure. Common forms, reciprocal reviews and shared secure environments can reduce duplicate work while preserving substantive standards.

Procurement strategy should also account for research uncertainty. Modular contracts, option periods and clear data rights can allow technical learning without locking an agency into one vendor or a prematurely fixed specification. Program officers and contracting staff need joint training because a scientifically elegant requirement can become unusable when contract terms prevent replication or access to underlying data.

A five-year implementation sequence

In the first year, the Office of Personnel Management, Office of Science and Technology Policy and major research agencies should define a common capability vocabulary and publish baseline workforce and hiring measures. Agencies should inventory shared facilities, critical datasets, software and projects at risk from single points of failure.

During years two and three, agencies should expand shared expert pools, technical rotations and infrastructure allocation while adding stewardship costs to grant and contract guidance. Major facilities should publish lifecycle and user-access plans. Pilot programs should compare new hiring and administrative pathways against normal time, retention and performance.

By year five, Congress and agencies should have enough evidence to decide which pilots deserve permanent authority. Public reporting should show whether critical gaps narrowed, access to infrastructure broadened, research outputs became more reusable and mission organizations adopted validated results. The review should also identify programs that added process without measurable capacity.

Implementation should begin with existing authority where possible and identify statutory barriers separately. This keeps administrative improvements from waiting behind debates that require Congress.

Tradeoffs and safeguards

Shared infrastructure can centralize authority or create dependence on private contributors. Workforce flexibility can weaken merit principles if exceptions become patronage. Open evidence can expose sensitive information. Continuity can become inertia. Translation metrics can pressure researchers toward short-term deliverables.

These risks require governance rather than abandonment. Allocation criteria should be public and appealable. Temporary hiring authorities should use documented qualifications and conflict rules. Data access should be tiered. Portfolio continuation should face scheduled independent review. Basic research should be evaluated on the knowledge and option value appropriate to its horizon, not forced into a near-term commercialization metric.

Limitations

This paper does not estimate the appropriation required for each recommendation or prescribe agency-by-agency staffing levels. Those decisions require current mission inventories, facility conditions and fiscal analysis. International R&D comparisons also depend on purchasing-power adjustments, classification and estimates that can be revised.

Capacity is not equivalent to scientific quality. A well-resourced institution can still select weak questions or resist correction. The proposed system must operate alongside peer review, research-integrity protections, ethical oversight and public accountability.

Conclusion

America’s research enterprise remains extraordinarily capable, but its public foundations cannot be inferred from the national spending total. Federal agencies supply basic-research support, facilities, data, standards and mission knowledge that private investment does not consistently replace.

Rebuilding public research capacity means making those foundations durable and usable: people who can exercise judgment, shared tools that broaden participation, evidence that survives the project and institutions able to learn across political and technological cycles.

Sources and methodology

This paper synthesizes federal statistical indicators, workforce audits, public-access policy and a current shared-infrastructure program. Recommendations are WSOI Institute’s analysis.

• National Science Board, The State of U.S. Science and Engineering 2026, executive summary.
• National Science Board, 2026 overview and R&D indicators.
• U.S. Government Accountability Office, Strengthening and Sustaining the Federal Science and Technology Workforce.
• National Science Foundation, National Artificial Intelligence Research Resource.
• Office of Science and Technology Policy, public access memorandum, August 25, 2022.