The US Government has made a series of rule changes aligned towards the priority on growing the country’s leadership in AI
Both the outgoing and incoming presidents have reprioritised investment in the US’ effort to lead global AI markets
Over the past two weeks, the US Government, under the leadership of both outgoing President Biden and incoming President Trump, has announced a range of executive actions aimed at boosting the country’s standing as a global leader in AI. Through a series of executive orders and new regulation, the Government has prioritised increasing domestic investment and further developing AI infrastructure, while limiting access to key inputs for foreign firms. As promised during the presidential campaign, President Trump has rescinded the prior executive order on AI safety, signaling a swift refocusing of US policy toward innovation and competitiveness in emerging technologies. While any of the other actions taken during the final days of President Biden’s term could also be rescinded, the outgoing president’s emphasis on supporting growth in AI markets as a matter of national security aligns well with President Trump’s intended approach and may result in some rules being maintained.
New rules will restrict exports of US-made AI chips and model parameters to a majority of the world
On 13 January 2025, the Department of Commerce’s Bureau of Industry and Security (BIS) issued its interim final rule on its Framework for the Responsible Diffusion of Advanced Artificial Intelligence Technology. The regulatory framework sets out a global licensing scheme for the export of advanced AI chips and the parameters (also referred to as weights) for advanced AI models. The BIS identified 18 countries, including Australia, Canada, Japan, South Korea, the UK and some EU Member States, as allied nations that represent a low security risk. As such, firms based in these countries will face no limit on the quantity of chips that they can import from the US. Another group of 23 countries were identified as prohibited countries, so any application made to export chips to firms based in these countries will be reviewed with the presumption of denial, which is likely to restrict all exports. All remaining countries will face an annual limit on chip exports based on a national calculation of total processing performance (TPP). The export of weights for closed AI models trained with 1026 or more computational operations will also be restricted to firms based in the US or the 18 whitelisted countries.
Data centre operators based in the US or in the 18 approved countries will be able to source unlimited quantities of chips to support the construction of data centres in any non-prohibited country around the world. Those operators will be subject to added security and reporting requirements but will not need advanced licensing. In total, these operators must maintain at least 75% of their total computing power within whitelisted countries, and US-based operators must maintain at least 50% of their computing power in the US. The framework has attracted extensive criticism from other countries as well as industry. US tech firms including Oracle and Nvidia have argued that the measure will allow Chinese competitors to gain traction in a number of key countries. The country classification system also notably excludes some NATO allies from favoured status as well as 17 EU Member States. In response to the US’ decision to treat EU countries individually, Henna Virkkunen (EVP for Tech Sovereignty, Security and Democracy) argued that the whole of the bloc “represents an economic opportunity for the US, not a security risk”. If President Trump chooses to maintain and finalise this new framework, firms will be required to comply by 15 May 2025.
The Biden Administration attempted to balance environmental concerns against infrastructure development before leaving office
On 14 January 2025, President Biden issued an executive order on AI infrastructure, committing additional federal land for private use in AI development. The order directs the Department of Defense and the Department of Energy to lease at least six federal property sites to private firms seeking to develop gigawatt-scale AI data centres through a competitive tender. In an attempt to balance environmental concerns, the order also requires data centre operators to cover all building and energy transmission costs while accessing needed power from clean energy sources. However, these projects will benefit from categorical exemptions from environmental reviews under the National Environmental Protection Act along with a prioritisation of speeding up permitting procedures. Though the order included conditions around interconnection with local energy grids that the Government claims will prevent a rise in retail energy prices, consumer and environmental groups have criticised the new measures as jeopardising decarbonisation goals and burdening local communities. If the executive order is not rescinded by President Trump, federal agencies will have until 28 February 2025 to identify the federally owned sites to be made available for data centre development.
The ‘Stargate’ joint venture between leading US firms will invest $500bn over four years to develop AI infrastructure
On 21 January 2025, President Trump joined leaders from OpenAI, Oracle and SoftBank to announce the formation of Stargate, a joint AI infrastructure venture between the firms. Stargate will invest $500bn (£407bn) over the next four years in constructing data centres in the US to support AI development, with $100bn (£81bn) of that coming immediately. President Trump estimated that the project would create at least 100,000 new jobs and described the investment as a "resounding declaration of confidence in America's potential". Arm, Microsoft and Nvidia will also join the project as technology partners, while the UAE’s technology-focused state fund MGX will be an added investor. Stargate’s first new data centre will be located in Texas, where construction is already underway, and the group is considering expansion to other states. Structurally, SoftBank will bear financial responsibility for the venture, and OpenAI will be responsible for operations. While the immediate source of the remaining $400bn (£325bn) is not yet known, this commitment nonetheless far outpaces investment levels in other countries and regions focusing on incentivising AI innovation, including the EU, Singapore, Taiwan and the UK.
