Why has the United States also begun to pursue "whole-of-nation innovation" in the AI era?

Five Institutional Shifts in "Science: A New Golden Age" Explained in One Article

On July 21, 2026, Michael Kratsios, Director of the White House Office of Science and Technology Policy, submitted "Science: A New Golden Age" to President Trump. The report, together with its appendices, totals 123 pages. Its cover, opening chapter and historical narrative are intentionally echoing "Science: The Endless Frontier" published 81 years ago.

In 1945, when World War II was drawing to a close, President Roosevelt asked Vannevar Bush to answer one question: How can the scientific research capabilities jointly formed by the government, universities and industry during the war be transferred to the peacetime period and continue to serve national security, health and economic development? Bush's answer laid the basic framework of the U.S. post-war scientific system: the federal government continuously supports basic research, universities become the main research sites, scientists conduct explorations in a relatively free environment, and enterprises then transform knowledge into technologies and products. The establishment of the National Science Foundation and the subsequent leading position of the United States in semiconductors, aerospace, the Internet, biomedicine and other fields are closely related to this system.

Kratsios now proposes that the system that helped the United States win the 20th century is no longer sufficient to cope with the 21st century. The problem is not that science is no longer important, nor is it simply that there is insufficient investment, but that the locations where science takes place, the ways to solve problems and the global competitive environment have all changed. Private enterprises invest about 700 billion U.S. dollars in R&D funds in the United States every year, far exceeding the total investment of the government and universities; the computing power, data, instruments and engineering teams required for many basic research projects have been concentrated in enterprises; scientific discoveries are increasingly dependent on large-scale collaboration, complex equipment and manufacturing feedback; AI has begun to enter the whole process of literature reading, hypothesis generation, experimental design and result verification.

Therefore, what this report discusses seems to be scientific research, but its deep-seated object is the system: after the technical conditions of science have changed, who will organize science, where the government will invest funds, how universities, enterprises and national laboratories will collaborate, and how achievements will be verified and transformed, all need to be redesigned.

I. What kind of report is this exactly

On March 26, 2025, Trump wrote a letter to Kratsios, asking him to answer three questions:

• How the United States maintains its leading position in key fields such as AI, quantum and nuclear technology;

• How to reduce administrative burdens and restore the vitality of scientific research;

• How to translate scientific and technological progress into economic growth and improvement of ordinary people's lives.

In the report, Kratsios further summarized the answers into four goals:

• Put scientists at the center of the scientific research system instead of traditional institutions;

• Change the distribution and evaluation methods of federal scientific research funds;

• Organize science around national goals and restore the "industrial muscle" that turns discoveries into industrial capabilities;

• Prepare the entire scientific research system for the AI revolution.

The main body of the report consists of five chapters.

• Chapter 1 explains why the post-war model needs to be updated;

• Chapter 2 discusses scientific research funding, peer review, administrative burdens and new-type institutions;

• Chapter 3 discusses national missions, regulation, public facilities and government-industry-university collaboration;

• Chapter 4 re-integrates manufacturing, skill training and regional innovation into the scientific system;

• Chapter 5 discusses AI, automated laboratories, scientific verification and the future publishing system.

The appendix is a memorandum on R&D budget priorities for Fiscal Year 2028 jointly issued by the White House Office of Science and Technology Policy and the Office of Management and Budget.

This appendix is very important. It means that the report is not just an ideological article, but also attempts to enter the federal budget and institutional implementation process. The memorandum requires federal agencies whose R&D budget reaches a certain threshold to submit action plans within 90 days, explaining how they will test new funding mechanisms, expand the openness of scientific research facilities, cultivate talents, reduce administrative burdens and strengthen connections with the industry.

To understand the entire report, we also need to separate the content of three different levels.

• The first type is policies and projects that have already been launched, including the "Genesis Mission", the "Gold Standard Science" executive order and the X-Labs of the National Science Foundation.

• The second type is budget and institutional reform guidelines for Fiscal Year 2028, whose realization depends on budget, departmental implementation and congressional authorization.

• The third type is future-oriented visions, such as "AI-native science" composed of AI Agents, scientific research bounties, automated laboratories and new credit allocation systems. These visions put forward directions, but there is still a distance from a mature system.

If these five chapters and the appendix are compressed into a more understandable map, five interconnected shifts can be seen. They are not five slogans listed word by word in the report, but five keys to understanding the report.

II. Why is the old scientific research system "no longer sufficient"?

The report first acknowledges that the post-war system still has irreplaceable value. Basic research has the characteristics of long cycle, scattered returns and strong spillover, and cannot be completely handed over to the market; national goals also require long-term coordination that transcends individual institutions. The report does not negate the principle of government support for basic research, but holds that "the principle still holds, and the process must change".

The first change is that the private sector has evolved from a receiver of scientific research achievements to an important participant at the scientific frontier. Taking the transistor and the Transformer architecture that modern machine learning relies on as examples, the report illustrates that enterprise laboratories not only engage in product development, but also raise basic questions, train scientific talents and create new research tools. In some fields, the computing power, data, instruments and remuneration mastered by enterprises are difficult for universities and federal agencies to replicate.

The second change is that the linear model of "basic research - applied research - product development" is increasingly difficult to describe reality. The material problems raised by semiconductor manufacturing will in turn promote fundamental physics, new scientific instruments will open up objects that could not be observed before, and experimental data will change the theoretical direction. Science, engineering, manufacturing and the market do not pass the baton in sequence, but iterate repeatedly.

The third change is that the investment in scientific research continues to increase, but it may not produce major breakthroughs of the same magnitude. Taking biomedicine as an example, the report discusses the rising R&D costs, the declining efficiency of drug R&D and the slowing growth of breakthrough achievements; at the same time, it points out that researchers often have to spend nearly half of their working time on applications, compliance and administrative affairs. From 1991 to the beginning of 2025, at least 270 new requirements were added to federal scientific research projects; some projects took up to 20 months from application to funding allocation. The report calls this consumption an invisible "innovation tax".

The fourth change is the simultaneous arrival of AI and national competition. AI can reduce the cost of raising hypotheses and processing knowledge, but it may make the bottlenecks in the old system more serious; global scientific and technological competition makes the United States no longer willing to assume that "as long as knowledge is discovered in the United States, the final benefits will naturally belong to the United States". As a result, scientific research policies have begun to pay attention to discovery, verification, manufacturing, supply chains and national security at the same time.

These four changes jointly promote the following five institutional shifts.

III. Shift 1: From funding institutions and projects to funding scientists and major issues

The traditional project-based system requires scientists to first write down the goals, routes, milestones and expected results, and then an expert committee decides whether to allocate funds. It is convenient for budget control and responsibility tracking, but it has a natural contradiction: real scientific exploration often cannot accurately state the results in advance, but applicants have to write the unknown into a seemingly predictable plan.

The report argues that the system should not only ask "whether this three-year project is written completely", but also ask "whether this person has the ability to raise important issues for a long time". Therefore, it proposes to expand postgraduate scholarships that can flow with researchers, increase independent support for young scientists, and refer to projects such as the "Director's Pioneer Award" of the National Institutes of Health to provide a small number of outstanding researchers with longer-term and less restrictive funding.

The so-called "funds follow people" does not only mean changing the collection account. It will weaken researchers' dependence on a single laboratory, supervisor and school, enabling them to choose a more suitable environment for the problem among universities, enterprises, national laboratories and independent institutions. The report especially hopes to shorten the time for young researchers to obtain independent status. It notes that the average age of new principal investigators in the U.S. biomedical field has been rising for a long time, and young scientists in their most creative stage often can only work for others' projects and are forced to choose safer directions that are easier to get funding.

"Funding major issues" means that the government no longer passively waits for applications from within disciplines, but identifies which scientific bottlenecks, once broken through, can open a series of subsequent discoveries. For example, a whole-brain connectivity map, general scientific instruments, sharable material databases or new experimental platforms may not be suitable for completion by a single laboratory, but can change the research capabilities of the entire field.

This shift can be simplified to understand: scientific research funds are not used to buy a prospectus, but to buy people's judgment, long-term concentration and freedom to change direction; the government not only funds an isolated research, but also can fund tools and capabilities that are commonly lacked in a field.

IV. Shift 2: From consensus-based peer review to diversified risk funding

The report does not advocate the abolition of peer review. It acknowledges that consensus review is suitable for eliminating obviously unqualified projects and for steadily advancing along mature routes. But when it becomes almost the only entrance for scientific research funds, its shortcomings will also be magnified.

Breakthrough research usually has insufficient evidence, unfamiliar paths, and even challenges the paradigms familiar to reviewers. In the case of a very low funding rate, what is most likely to pass is often not the plan that "a small number of people strongly believe in", but the plan that "most people do not strongly oppose". What is formed as a result is not an obvious error, but a systematic conservatism.

The approach proposed in the report is to configure different selection mechanisms for different problems.

The "Golden Ticket" allows a reviewer with technical judgment to strongly recommend an unconventional research that cannot obtain committee consensus; fast funding reduces application materials and shortens the decision-making cycle; long-term personal funding allows researchers not to rewrite their proposals every few years; secondary funding gives part of the choice to scientists who are closer to the frontier; scientific research bonuses and pre-market commitments do not specify technical routes in advance, but set rewards for verifiable results; ARPA-style project managers can actively find teams, adjust combinations and terminate invalid routes around a goal.

Behind these tools is a "portfolio" mindset. The failure of a single high-risk project does not necessarily mean the failure of the funding mechanism; a more important issue is whether a group of projects has exchanged a few truly important breakthroughs at an affordable cost. The federal government allocates about 200 billion U.S. dollars in R&D funds every year. The report hopes that agencies will not only allocate funds in accordance with the law, but also act as serious capital allocators to explain why their portfolios include different risks, durations and organizational forms.

In order to prevent the "new tools" from solidifying into old procedures again, the report proposes to establish a high-level meta-science unit within scientific research institutions, which is a very important new concept. Meta-science does not study a certain natural object, but studies "how science operates": compare the effects of different review methods, track the long-term performance of funded persons and shortlisted candidates, identify which fields have duplicate investments and which public tools have not been built for a long time, and adjust the next round of policies accordingly.

Therefore, this shift is not only about inventing "Golden Tickets" or fast funding, but also requiring the funding system itself to become an object that can be tested, evaluated and corrected.

V. Shift 3: From the university-centric model to a multi-stakeholder scientific research ecosystem

When the U.S. post-war scientific research system was formed, universities had the most concentrated knowledge, talents and equipment. Today, universities still undertake free exploration, talent cultivation and knowledge inheritance, but no longer have all key resources. Large technology enterprises can invest huge computing power and data, national laboratories have particle accelerators, light sources, supercomputers and nuclear facilities, start-ups are good at rapid engineering, and charitable funds can support long-term projects that are difficult for both the government and the market to undertake.

At the same time, many problems fall into the gaps of existing institutions:

• Too large for a laboratory led by a university professor, and too complex for a single discipline;

• Too slow in return and too public for enterprises;

• Too flexible or short-term for permanent national laboratories.

The report therefore proposes to expand the types of "implementers" of scientific research. ARPA-style institutions organize tasks by project managers with large discretionary power; focused research organizations consist of ten to one hundred professional scientists and engineers working around clear bottlenecks for several years; curiosity-driven research institutes provide a long-term stable environment for a small number of researchers; X-Labs use independent, full-time, milestone-driven teams to build platform technologies that can open up new fields.

X-Labs is not a paper vision. In May 2026, the U.S. National Science Foundation announced that it plans to invest up to 1.5 billion U.S. dollars in ten years, with the first round focusing on quantum system interconnection, integrated photonics, and new-generation scientific sensing and imaging instruments. It hopes to fund independent teams that are difficult for traditional universities and enterprise laboratories to undertake, and allow teams to advance their achievements from concepts and prototypes to platforms that can be further scaled up by private capital.

The report also hopes that talents can flow more smoothly between different sectors, including industry doctoral and postdoctoral programs, federal laboratory internships, joint centers between enterprises and universities, and stable positions for professional scientists and engineers. This means that researchers do not have to regard tenured faculty positions as the only path to success, and scientific capabilities no longer only exist in university departments.

This shift is not to replace universities with new institutions, but to match organizational forms with problem forms. Small-team free exploration, mesoscale platform construction, major national missions and commercial product development inherently require different time scales, talent structures and accountability methods.

VI. Shift 4: From linear achievement transformation to the cycle of science, engineering and manufacturing

The report's understanding of achievement transformation is broader than "applying for a patent after publishing a paper and then licensing it to an enterprise". It believes that real technical capabilities not only exist in papers, patents and prototypes, but also in the tacit knowledge formed by engineers and technical personnel through long-term practice.

To turn a laboratory discovery into industrial capabilities, it still needs to go through prototyping, testing, pilot scale-up, regulatory verification, engineering scaling, supply chain adaptation and first market procurement. If any link is missing, knowledge may remain on paper. More importantly, these links are not auxiliary work after the discovery is completed. Manufacturing defects will expose new scientific problems, instrument capabilities will determine what scientists can observe, and engineering scaling will also force researchers to re-understand materials and processes.

The report refers to this relationship as the combination of science and "craftsmanship". Advanced manufacturing is not just about more machines and higher automation, but also relies on process personnel who know how to debug equipment, handle anomalies and control yield. A lot of knowledge cannot be fully written into documents, and can only be accumulated among mentors and apprentices, teams and production sites. This kind of knowledge is tacit knowledge. In his book "Tacit and Explicit Knowledge", British sociologist Harry Collins divides tacit knowledge into three basic types: relational, bodily and collective. When manufacturing links are moved outwards, what is lost is not just jobs, but also knowledge, including the problems, skills and supply chain experience required for the next round of innovation.

Therefore, the report advocates opening federal facilities and test sites such as the U.S. Department of Energy's national laboratories and NASA centers to enterprises and non-traditional