Last month, the release of the RI model by the Chinese artificial intelligence (AI) company, DeepSeek, shocked the US tech industry. Yujia He writes that while DeepSeek’s model may not be quite as superior compared to its competitors as originally thought, the Trump administration has responded to its innovations by further pushing for US leadership in AI. Taking a close look at US research and development policy under the second Trump administration, however, shows the significant disconnect between rhetoric and policy. Federal funding freezes and reduced funding for the National Science Foundation could have dire long-term consequences for research and international cooperation including on AI.

The launch of an AI chatbot in late January based on the latest reasoning model from Chinese startup DeepSeek sent shockwaves in the US tech industry. Silicon Valley players were quick to weigh in: venture capitalist Marc Andreessen called DeepSeek “AI’s Sputnik Moment”. As investors learned about this model reportedly built at less cost and using inferior chips compared to those from US-industry leaders, tech stocks crumbled. Nvidia, the leading specialized chip maker for building AI systems, saw a $600 billion loss in its market value in one day.

DeepSeek: Was the hype justified? 

Now that the tech stock crash has abated, new analysis has cut down some of the hype over DeepSeek: according to MIT Technology Review, instead of the $6 million development cost initially reported by the press, the cost of research that led to Deepseek’s innovation is likely to be 10 times or higher than its rivals (though still less than most American models). Initial hope about energy efficiency has been replaced by sobering projections that Deepseek’s reasoning model may be just as energy intensive: “the energy it saves in training is offset by its more intensive techniques for answering questions, and by the long answers they produce.” There are various analysis and speculations about DeepSeek’s access to high-end AI chips, and what its success means for the efficacy of complex export controls set by the Biden administration. On this issue, some argue for continuing and expanding export controls on AI chips, while others counter that export controls are based on key mistaken assumptions.  

Still, DeepSeek seems to have brought in new innovative developments. The model uses reinforcement learning, a machine learning technique, to enhance the capability of its reasoning, removing steps that required human feedback while optimizing the use of less high-end chips. Unlike OpenAI which made its model proprietary, DeepSeek published the details in a paper and have released their model as open source, so outside researchers and developers could see how it was done and train their own reasoning models to solve problems (as an extreme example, UC Berkeley researchers developed a small-scale language model reproduction of DeepSeek useful for very restricted types of tasks for $30). There are concerns about censorship and data privacy using DeepSeek’s online chatbot, which stores user data in servers in China. Users can get around these by downloading and using a locally-hosted version independent of DeepSeek and modifying it, or by using services created by other companies that have integrated the DeepSeek model. It seems that with DeepSeek’s associated cost reduction more business enterprises may build and eventually deploy AI applications at scale.

AI takes center stage in the new Trump administration’s science and technology policy 

The DeepSeek frenzy may have reinforced the new Trump administration’s purported policy priority to enhance America’s AI leadership. Donald Trump commented that DeepSeek “should be a wake-up call for our industries that we need to be laser focused on competing”. Over thirty cabinet-level departments and federal agencies have a role in funding research and influencing science policy, and many agencies now face uncertainties in agency leadership and funding, so there are many unknowns about the administration’s policy agenda. Still from Trump’s early picks for key White House science advisory positions, one can see a strong affinity for AI and for tech capital interests.

Within the White House, the Office of Science and Technology Policy (OSTP) coordinates science and tech policy across the federal government. Michael Kratsios serves as both its new director and the President’s science advisor (a cabinet-level position). While lacking a research background typically found in White House science advisors, Kratsios was previously Chief of Staff to venture capitalist Peter Thiel and the Chief Technology Officer at OSTP during Trump’s first term. During the Trump 1.0 administration, OSTP championed the “industries of the future” like AI and quantum computing, emphasizing AI-related R&D, accelerating AI adoption in the federal government, avoiding stringent industry regulation, and keeping US leadership over China. As a comparison, climate change and pandemic response were virtually absent from OSTP’s agenda. Trump 2.0’s OSTP under his leadership is likely to have a similar trajectory.

The President’s Council of Advisors on Science and Technology (PCAST) is a key advisory body composed of external experts like university researchers and industry leaders. PCAST under this administration is chaired by David Sachs, a tech investor and former PayPal executive known as part of the “PayPal Mafia” including Peter Thiel and Elon Musk. Sachs has also been named as the White House “AI and crypto czar”. In a media interview about DeepSeek, parroting the administration’s focus on deregulation and ending DEI (Diversity, Equity and Inclusion), Sachs blamed the possibility that DeepSeek stole American intellectual property, and criticized the Biden administration’s regulations and American tech firms’ DEI efforts for producing “woke AI”. He also defended the Trump administration’s support of Stargate, a private-sector joint venture investment of $100 billion (eventually $500 billion) by OpenAI, Oracle and SoftBank to build large-scale data centers and energy infrastructure across the US.

The disconnect between rhetoric and policy: Uncertainties for international cooperation and federal support for R&D 

For the research community however, the administration’s purported focus on advancing American leadership in AI stands in sharp contrast with its chaotic handling of federal support for research and development (R&D). Only seven days after the inauguration, federal funding from the National Science Foundation (NSF) and the National Institute of Health (NIH), two of the largest funding sources for university research beside the Department of Defense, were abruptly halted following an Office of Management and Budget (OMB) order to halt all federal grants. Agency employees were also told to suspend all meetings and travel, including grant reviews and advisory committee meetings. While the funding freeze was temporarily lifted by a court order, the administration has countered that existing funds be frozen until reviews are completed to ensure compliance with the President’s various executive orders, such as ending DEI support. Yet the NSF follows a 1980 Congressional mandate to broaden the participation of under-represented groups such as women and people of color in scientific research, meaning that the agency and its funded research are now caught in the limbo between the mandate and the administration’s requirements. The funding freeze has led to significant chaos across universities and research institutions, as researchers struggled for clarity over what to do with pending applications and existing grant renewals, and with the NSF online payment system offline, transactions such as equipment purchases and salary payments were stalled. In addition to the funding fiasco, researchers globally have scrambled to save data from American government agencies, as many databases and resources have disappeared from data.gov and agency websites.

Photo by Saradasish Pradhan on Unsplash

The federal budget typically follows a complex political process that involves bargaining between the White House and Congress every year, with the judicial branch also playing a role in determining the legality of policies, so there are political battles in the months to come. As an example, despite the 2022 CHIPS and Science Act authorizing $200 billion spending on R&D in the next decade, including for the NSF to double its budget between 2023 and 2027, Congress is not obligated to fund it at those levels, and the NSF budget has remained low at about $9 billion. Yet the administration’s proposal for the next fiscal year starting October 1 plans to slash the NSF budget to as little as $3 billion. The final budget approved by Congress may be different. Still, considering that the NSF pays for 80 percent of federal funding for computer science in US academic institutions, one wonders how the administration aspires to maintain the US lead in AI, when it simultaneously gives such a chaotic and bleak funding outlook for university researchers and educators in the field.

Consequences of the federal funding freeze for US and international research

And there will be long-term consequences if the broader trend continues. Federal support funds more than half of American universities’ R&D expenditure and has multiple functions: it pays for direct research expenses, student support, and some of the indirect costs that fund universities to cover associated expenses like buildings and operating facilities. Facing uncertainties in public funding, university researchers may choose to pivot away from early-stage research that the private sector has less interest in supporting. Students may choose to forgo further education opportunities, leading to a shrinking pipeline of talent for both research and industry careers. Becoming a scientist in government agencies may no longer be a desirable or feasible career option, leading to a decline of scientific expertise for federal policymaking. Universities are often the major employer in many college towns across the US, thus a decline in federal support for research and education would also negatively impact local economies.

Internationally, the funding uncertainty, coupled with uncertainties over visas and immigration, may dissuade researchers and students from moving to US universities for research and study. When US federal agencies face uncertainties in budgets, employee travel and international communication, international knowledge exchange, regulatory cooperation and experience sharing among government agencies in charge of science and tech-related missions will be impacted. This is already happening at the National Oceanic and Atmospheric administration (NOAA), which has been ordered to halt all “international engagements” including international travel, meetings and emails with foreign national colleagues, despite the fact that hurricanes and marine life pay no attention to international borders.

The AI race and the Trump administration

There is also increasing uncertainty on the US’ role in international cooperation in AI governance. The Trump administration has revoked an earlier executive order by the Biden administration to establish standards for AI safety, promote responsible innovation and public-sector AI deployment, protect against unlawful discrimination and bias, support workers affected by AI-driven changes, and collaborate with other nations to establish AI safety benchmarks, promote ethical AI deployment, and address cross-border challenges such as cybersecurity. This means that some of the ongoing work at agencies such as the National Institute of Standards and Technology (NIST), which houses the US AI Safety Institute and monitors AI development, might be curtailed or shifted in new directions.

Meanwhile, the frenzy around DeepSeek and Chinese AI advancement more broadly appears to be useful in the push for a stronger confluence of tech capital interests with the administration’s policy agenda, with the “AI arms race/cold war” narrative gaining further momentum. A day after Trump was inaugurated, the CEO of Scale AI, a company that manages freelance contractors to do data labeling for training AI tools for firms like Meta and Google and for the Department of Defense, took out a full-page advertisement in the Washington Post lobbying for greater support from the President that “America must win the AI War” against China. Interestingly, Michael Kratsios who heads the current OSTP, was a managing director of Scale AI after leaving the CTO job in the Trump 1.0 administration. The company has also been sued for multiple allegations of labor violations.

Time will tell if these recent developments are temporary, and what long-lasting impacts they may have. If the US research enterprise, and the policy institutions supporting it, starts to fracture from within, it will not be good for the progress of science and technology for the entire world.

• Subscribe to LSE USAPP’s email newsletter to receive a weekly article roundup.
• Please read our comments policy before commenting. 
• Note: This article gives the views of the author, and not the position of USAPP – American Politics and Policy, nor the London School of Economics. 
• Shortened URL for this post: https://wp.me/p3I2YF-eWI