On his first day in office, President Donald Trump repealed Biden’s “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence” executive order, signaling a new policy approach at a pivotal moment for AI progress. Now, reports suggest that the Trump Administration may eliminate the AI Safety Institute.
As researchers project the possibility of human-level AI within the next decade, America’s leadership position is precarious. China is working aggressively to narrow the capability gap, investing massive resources and undertaking covert actions to bypass export controls and to steal sensitive AI secrets.
While many believe it to be overhyped, Chinese company DeepSeek’s latest AI model demonstrates China’s ability to quickly follow the U.S. lead. The Trump Administration now faces a critical challenge: how to maintain and strengthen American leadership in AI while protecting its advances from foreign adversaries.
The most immediate challenge is infrastructure. Improvements in more efficient models, like DeepSeek’s recent breakthrough, don’t change the fundamental reality that training, using, and deploying the most capable AI models at scale will require massive amounts of computing power and energy. Within five years, training a leading AI model could require the equivalent of five nuclear reactors worth of electricity. Yet America’s aging power grid and byzantine permitting processes can create significant delays in constructing the energy sources needed to develop and power the most cutting-edge AI models. While some U.S. firms are exploring breakthrough solutions like small modular nuclear reactors, the regulatory hurdles remain daunting.
Government action and private investment are converging to address these bottlenecks, but they remain woefully insufficient. One of President Biden’s final executive orders laid the groundwork to lease federal land for gigawatt-scale AI data centers with expedited permitting. President Trump has used his first days in office to go even further. He declared a National Energy Emergency that explicitly highlights the imperative of securing America’s lead in technology innovation and weakened rules under the National Environmental Protection Act to help fast track deployment of new energy sources nationwide. Following this step, President Trump joined the leaders of OpenAI, Softbank, and Oracle to announce the Stargate Project–a $500 billion private initiative to build new AI and energy infrastructure in the United States.
But even these dramatic efforts will not be enough. Fundamental challenges persist in modernizing the power grid and coordinating federal, state, and local authorities. In a shifting and contested regulatory environment, lawsuits and court rulings could produce multi-year delays. Washington still requires a comprehensive strategy that goes beyond reversible executive actions to address these deeper policy and regulatory headwinds over the long-term. Without broader reforms and sustained coordination across government, America’s AI leadership will face continuing constraints.
Security must underpin President Trump’s AI agenda. There’s no point in investing billions of dollars in U.S. AI infrastructure if the Chinese Communist Party can simply swoop in and steal the AI models after development. Developing cutting-edge AI models requires multi-billion dollar investments in specialized chips, datacenters, energy, and top talent, and those costs will only grow over time. The Biden Administration imposed export controls to block the sale of advanced chips and chip-making equipment to China in a bid to slow its AI progress. If China and other competitors can simply steal the next generation of trained AI models, they will have effectively bypassed the controls to benefit from multi-billion dollar U.S. investments. That would be a colossal self-inflicted wound. As AI capabilities advance and training costs escalate, the gap between the U.S. and China will continue to grow, creating even stronger incentives for rival nations to steal these technologies. This also raises critical questions around if, and when, open release of model weights—the sensitive intellectual property that encapsulates the capabilities of a trained model—should be restricted. It is crucial for the United States to secure its AI advantages against theft and exploitation. The Trump Administration has an opportunity to work with the private sector to set risk-proportionate security standards for leading AI developers and data centers. Only then can we ensure that America benefits from its unrivaled investment and innovation.
Efforts to build out and secure U.S. AI infrastructure will require sustained efforts over the long-term. This reality, however, has created a short-term challenge: American companies are now seeking to build cutting-edge AI infrastructure overseas in places like the United Arab Emirates and Saudi Arabia that offer abundant energy and limited regulation. We should promote the diffusion of American AI to reinforce U.S. technology leadership and preempt China’s efforts to establish a Digital Silk Road for AI.
At the same time, American AI companies’ short-term demand for energy and capital could, absent government involvement, lead to powerful AI infrastructure being built abroad, which could potentially result in the most advanced models being developed beyond U.S. control and more vulnerable to Chinese espionage. The Trump administration should not accept this.
Instead, it should develop a comprehensive strategy for spreading both U.S. AI hardware and software globally, while protecting and reinforcing America’s advantage. It can start by building on the Department of Commerce’s Framework for AI Diffusion. As proposed in January 2025, the framework controls the global spread of advanced AI chips by splitting the world into three tiers: one for close allies and partners, who have unrestricted access, another for geostrategic rivals such as China, who have no access, and a third for the rest of the world, who can access advanced AI chips up to certain limits. However, the framework privileges American companies building overseas, enabling them to access unlimited numbers of advanced chips provided they adhere to strong security standards and keep 50% of their total AI computing power in the U.S. This framework also provides a pathway for non-U.S. companies to build large datacenters overseas with a relatively high AI chip limit, once they have been determined to meet security standards and divest from Chinese companies. In addition, the framework introduces security protocols to protect non-public AI model weights.
While this framework is a strong first step, its emphasis on export controls has cast it as primarily restrictive, creating openings for Chinese influence. Though chip restrictions remain vital, the U.S. must balance these controls with an aggressive strategy to spread other parts of AI technology, ensuring American innovation underpins the global ecosystem. Just as U.S. dominance of the financial system allowed it to set global rules and enforce sanctions, being central to global AI deployment will be key for American strategic influence.
Moving from a restrictive to proactive strategy, President Trump can turbocharge U.S. AI leadership internationally by making AI deals with strategic partners. For example, India was withheld from the top tier in the framework but is a key technology partner for the United States with large AI ambitions. As proposed, the framework creates friction in the very U.S.-India tech cooperation that will be vital to countering China’s global tech influence. A more comprehensive government-to-government approach should extend beyond chip access to encompass cloud computing, AI model partnerships, and downstream applications—from drug discovery to manufacturing—to demonstrate the benefits of partnering with the United States on AI. Through targeted bilateral engagement and bespoke deals that adapt to different partners’ AI needs and level of risk, President Trump has an opportunity to tailor partnerships that strengthen U.S. influence while helping partners access AI capabilities securely.
While ensuring the United States leads the frontier of AI development, the Trump Administration should actively monitor for dangerous capabilities and be ready to prevent them from being exploited by adversaries. Such capabilities are most likely to emerge from the most advanced models which are trained on growing amounts of computing power.
Rather than entirely gut the AI Safety Institute, President Trump should instead refocus its talents towards the most critical national security capabilities and risks, ensuring they are identified if they arise. With leading labs already partnering closely with national security agencies, there are growing avenues through which to responsibly assess and manage such capabilities.
In cases where significant risks are identified and can’t be mitigated by safeguards, it will be essential that the U.S. government has the ability to restrict their release to prevent them from being used by adversaries against the United States. There should be a high bar for any such intervention, as a vibrant open-source system is vital for innovation and a key enabler of the U.S. technology sector. Striking the right balance is critical. The new AI agenda must not be about restricting AI development, but rather about ensuring America’s technological leadership develops in a way that benefits U.S. interests.
The next few years will be decisive. The nation that solidifies its AI advantage will shape the trajectory of the most transformative technology of our era. America has led the world’s technology revolutions before—from the assembly line to the nuclear age to space, semiconductors, and the internet. It can lead the AI revolution too, but only with urgent action to promote and protect American advantages.
The post As Trump Reshapes AI Policy, Here’s How He Could Protect America’s AI Advantage appeared first on TIME.
