Key points
The U.S. Department of Commerce and other regulatory agencies released regulations at a frenetic pace as the Biden administration wrapped up its term this month, as The new "Export Control Framework for Artificial Intelligence (AI) Diffusion".
The Framework for the diffusion of artificial intelligence was published on January 15, 2025. BIS has implemented significant additional controls on advanced computing chips and certain closed AI model weights, as well as new licensing exceptions and updates to validated end-user (VEU) permissions. The framework adopts a “three-pronged strategy” that i) requires authorizations for exports, re-exports and transfers of advanced node ICs to a broad range of countries; (ii) establishes new controls on the weights of more advanced AI models, including a new FDP rule that applies such controls to certain weights of models produced abroad using advanced computer chips made with U.S. technology or equipment; and iii) imposes additional security conditions relating to the preservation of certain AI models. While the rule is effective immediately, the expected compliance date for these rules is set for May 15, 2025, with some elements of these controls having a delayed compliance date of January 15, 2026.
Biden's restrictions on the export of AI chips aim to prevent the technology from being misused. However, large companies fear high costs and market loss, while excluded countries express concern about limited access to advanced technologies.
These new regulations appear primarily aimed at limiting China's ability to further develop certain critical technologies, such as artificial intelligence (AI), or access the US market, but each has a broader impact and could have some unintended consequences for industries hit.
It remains to be seen whether and how the implementation of these rules will be reviewed in the coming months, particularly in light of President Trump's "Day One" Executive Orders (EOs) implementing a regulatory freeze and reversing many actions taken by the Biden Administration .
To the extent that these rules impose restrictions on China, an area where the two administrations have historically been largely aligned, they are more likely to be maintained under the new administration. Other rules can be canceled or changed.
We expect to issue more in-depth notices as we learn more about the new administration's plans for these rules. In the meantime, we recommend that companies consider the potential impact of the rules and be prepared to implement any necessary changes.
One of the latest actions adopted by the Biden Administration is the recent Interim Final Rule (IFR) for a new "Export Control Framework for Artificial Intelligence (AI) Diffusion" which, however, is sparking a heated debate in the technology sector.
To address national security concerns, including the potential misuse of AI to develop weapons of mass destruction or enable malicious activities, this regulatory framework developed by the Bureau of Industry and Security (BIS), an agency of the US Department of Commerce United States, imposes strict licensing requirements for exports of AI technology.
In particular, this rule: i) places country-specific caps, together with, as mentioned, a licensing regime on semiconductor exports that would have a particular impact on GPU chips that underpin key AI applications, such as the development of large language models (LLM); ii) determine where to ship essential American AI chips; iii) determines where to build data centers that create AI, with a preference for the United States and its allies.
However, major national suppliers including Oracle, Microsoft, Amazon and Meta argue that this regulatory framework is excessively pervasive as it would impose excessive burdens, limit innovation and consequently favor the cession of market leadership to China.
“Not only does it limit China's access to advanced technologies, pushing US companies to innovate and develop alternatives; it also limits the global market share of US companies and encourages China to accelerate its own technological advances, altering the global technological landscape and intensifying technology competition between the United States and China,” said Charlie Dai, vice president, principal analyst at Forrester.
OBJECT OF THE STANDARD
The new one Interim Final Rule (IFR), which is over 200 pages long, provides for the establishment of a mandatory global licensing requirement for AI technology and GPUs (GPUs are electronic circuits capable of performing mathematical calculations at high speed) which is linked to specific limits for Country on US exports of these technologies. Twenty trusted countries (Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, the Republic of Korea, Poland, Spain, Sweden, Switzerland, Taiwan and the United Kingdom) would be exempted from export restrictions. But for the more than 140 non-exempt countries (a list that includes some of America's most important allies like Israel and Singapore; key trading partners like Brazil, India, Indonesia, Malaysia, and Mexico; and other allies like Saudi Arabia and the United Arab Emirates ), the regulations create a Low Processing Performance (LPP) license exemption, essentially allowing certain levels of GPUs to be shipped to these nations on a limited, country-specific basis. To handle the challenge posed by U.S. companies sending AI systems to their subsidiaries operating in foreign countries or to the data centers they manage, the regulations would establish a regime Universal Validated End User (UVEU) which would allow trusted partners (such as US hyperscalers) to receive faster access to GPUs, but only if they comply with federal FedRAMP High requirements (Federal Risk and Authorization Management Program) a technical certification designed to protect the federal government's most sensitive unclassified data. In particular, it is about a government-wide program that provides a standardized approach to ongoing security assessment, authorization, and monitoring for cloud products and services. In 2022, Congress codified FedRAMP as “a government-wide program that provides a standardized, reusable approach to security assessment and authorization for cloud computing products and services that handle unclassified information used by agencies.”
EFFECTS ON BUSINESSES
The implications for businesses are far-reaching. Under the new regulatory framework, companies that rely on GPUs for cloud services or AI development could experience cost increases, supply chain issues and delays in accessing cutting-edge technologies. For cloud providers, the compliance costs of adapting data centers to stringent security requirements could create additional burdens.
“For businesses, the export control regulatory framework could disrupt GPU supply chains, causing project delays and increased operational costs, and forcing businesses to invest in alternative technologies, potentially impacting their competitiveness and profitability "Forrester's Dai said.
Cloud providers “would face challenges in complying with complex regulations, risk losing market share as customers seek alternatives, and could face limitations on innovation due to the diversion of resources towards regulatory compliance,” he stated.
While the framework aims to strengthen oversight, critics fear it risks straining international partnerships and stifling global AI collaboration.
Oracle has criticized the framework in a blog post, saying it will disrupt U.S. leadership in cloud, chips and AI, instead of focusing on targeted, high-risk businesses.
“The Framework introduces so many new acronyms into what we might best call the Confusion Framework that it's hard to keep them in mind: AIA, ACM, LPP, DC VEU, UVEU, NVEU, TPP, ACA. The Framework identifies 20 Artificial Intelligence Authorization Countries (AIA) that enjoy modestly better regulatory treatment than the rest of the world, but at the same time creates a regulatory quagmire for cloud providers to even serve some of our closest allies. In one confusing action, BIS retroactively regulates global cloud GPU deployments; narrows the global market for US cloud and chip providers; establishes volume restrictions; tells 20 countries that they can only be trusted if they accept new unilaterally imposed terms - including certification and semi-annual reporting requirements - and likely pushes the rest of the world towards Chinese technology, which the CCP will be only too happy to exploit to catch up with the United States,” Oracle said.
“The main problem with the deployment framework is that the global commercial cloud has been built continuously and globally over the past twenty years. Large investments have been made. Commitments have been made to customers. Location decisions are driven by infrastructure such as energy and bandwidth. Many critical questions appear to go unanswered or not even considered before an IFR is issued. How does the rule fit with sovereign clouds deployed around the world with prior authorization from the US government? What about regulated customers, like banks, who deploy cloud in their own data centers? And a national health system? Does the technology refresh count towards the national caps? What about data centers co-located and managed by others? Will existing data centers all need to meet US government requirements of UVEU FedRAMP High?” Oracle added.
“What Congress accomplished with the $280 billion CHIPS Act is being undermined, with this rule slashing the global chip market for US companies by 80% and handing it over to Chinese competitors,” Ken wrote in the blog Glueck, executive vice president of Oracle. According to him, the regulatory framework would be more appropriately called “Export Control Framework for the Advancement of Alibaba, Huawei, Tencent and SMIC.”
An Nvidia spokesperson declared that the new rule would threaten U.S. economic growth and leadership. “Every data center is being accelerated, and every business and application is incorporating mainstream AI. The worldwide interest in accelerated computing for everyday applications is a huge opportunity that the United States can cultivate, boosting the economy and adding jobs to the United States." A last-minute rule limiting exports to most of the world would be a major policy shift that would not reduce the risk of misuse, but would threaten economic growth and U.S. leadership.”
There were strong criticisms also from the Information Technology and Innovation Foundation (ITIF). According to the Vice President for Global Innovation Policy Stephen Ezell, “the proposed regulations misunderstand a key element of how AI chips are used to develop LLM and work on other computing challenges: GPUs scale both up and down, meaning their capability comes from operating simultaneous use of many GPUs to address a computing challenge. Therefore, GPU control makes little sense when a competitor can achieve parity by simply adding more, albeit less powerful, GPUs to solve the computing challenge. In other words, even if Chinese-made GPUs don't perform as well as US ones in the short term, Chinese companies (or those of other competitors) will be happy to provide the computing power needed to address customer challenges, thus frustrating the administration’s goal of limiting global AI computing while damaging U.S. companies’ leadership in global AI computing technology and market share.”
“Furthermore, the regulatory framework would introduce significant, likely intractable, compliance challenges. First, it would be extraordinarily difficult (if not nearly impossible) for companies to know whether exporting a product would put a country above the count limit, unless there is an onerous process under which an entity U.S. government examines all sales and aggregates sales information from the entire industry (including those of foreign companies). The current AI chip export authorization process, which only applies to some countries (and China), is also unpredictable, expensive and time-consuming; therefore, extending this process globally would be highly impractical, and this would be true for any country-level upper limit, regardless of how high,” Ezell added.
“The Biden administration is trying to force other countries to choose sides - the United States or China - and will likely find that if it issues this ultimatum, many will choose China. After all, only one of these countries is actively threatening to cut them off from the precious AI chips they will need to compete in the digital economy. The next administration should focus on improving U.S. competitiveness in the AI sector by expanding market access for U.S. AI chips and technologies and limiting the influence of its geostrategic competitors in the sector. The vast majority of uses of these chips will be for legitimate and lawful purposes, and the goal should be to capture and retain as much of this market as possible. Additionally, the United States should counter efforts by China and Russia, which recently launched a network of AI alliances among the BRICS countries, to give their allies and partners access to AI data, models, and computing resources. 'AI,” concluded the ITIF.
Also the Information Technology Industry Council (IN), a powerful lobbying group representing industry giants like Amazon, Microsoft and Meta, has issued a stark warning. In a letter addressed to Commerce Secretary Gina Raimondo, ITI CEO Jason Oxman condemned the administration's eleventh-hour push to finalize the rule, arguing that it would impose arbitrary limits on American companies' ability to engage in foreign markets. “Hastening the completion of a complex and consequential rule could have significant adverse consequences,” Oxman said in the letter obtained from Reuters. The ITI emphasized that while it recognizes the importance of national security concerns, “the potential risks to US global leadership in AI are real and must be taken seriously.”
Even the Semiconductor Industry Association (SIA) released a statement expressing reservations, further amplifying the industry's growing unease. At the heart of the Administration's initiative is a strategic goal: to prevent advanced artificial intelligence technology from strengthening China's military capabilities. “ASI and its member companies share the U.S. government's commitment to safeguarding national security. However, we are deeply concerned by the unprecedented scope and complexity of this potential rulemaking, which was developed without input from the 'industry and could significantly undermine US leadership and competitiveness in semiconductor technology and advanced artificial intelligence systems,' the report reads. “We respectfully object to such a rapid and significant change in policy in this transition period and without meaningful consultation with industry.” In the absence of such consultations, we urge the Administration to issue a proposed rule or delegate the policymaking process to the incoming Trump Administration to ensure adequate opportunity for government and industry leaders, together with our global partners to thoughtfully address this crucial issue."
“It appears that AI is about to become more explicitly nationalist and foreign policy aligned. This could accelerate AI sovereignty initiatives already underway.
Think of this as an analogous walled garden versus open-source discussion,” ha declared Abhishek Sengupta, Practice Director at Everest Group, noting that the move could further align AI developments with domestic and foreign policy agendas. Sengupta warned that while the United States can leverage AI diplomacy to attract new allies, it risks losing others to emerging global AI alternatives. “In the long term, this could have a serious negative effect on the competitiveness of US-based AI offerings,” he added.
UNWELCOME EFFECTS
Restrictions on access to high-end chips could also push adversary nations to innovate more efficient algorithms, Sengupta noted, citing China's DeepSeek as an example. “DeepSeek relied on NVIDIA's H800 chips, which have lower performance than NVIDIA's cutting-edge offerings limited to the Chinese market. The result? A more compute-efficient LLM that surpassed popular industry names like ChatGPT -4o and Llama 3.1 in several benchmarks and that cost only $5.5 million,” he explained.
“Establishing a system of globalized national limits for AI computing is as bad an idea as the US government's idea of controlling food (or pharmaceutical) prices,” he said. declared Ezell. “It is also clear that the Biden administration is trying to impose these rules in the final days of its term, preventing the next administration from assessing the central challenge and how to respond to it. The Export Control Framework for AI Deployment represents a flawed policy that should be immediately withdrawn and replaced with a better approach in the next administration, based on broader consultation and stakeholder input.”
THE SUPPORTERS
For supporters of the regulation, controlling AI exports is a necessary step to prevent misuse by foreign adversaries. The administration cites China's ambitions in AI and semiconductors as the driving force behind the rule: "We must play hardball with our adversaries on AI," Brad Carson, president of the non-profit organization, said in a statement. profit Americans for Responsible Innovation. “The Chinese Communist Party will exploit any loophole to get its hands on AI technology, so it is critical to stop the flow of this technology to China and its military allies. Big Tech will complain about the new regulatory framework, but ultimately Stopping the rapid advancement of AI technology to U.S. adversaries is a national security imperative.”
The regulatory framework reflects the Biden administration's ongoing efforts to stymie China's technological development in semiconductors and artificial intelligence, and comes just weeks before the expected swearing-in of President-elect Donald Trump, who is expected to maintain a tough stance on China.
Effectiveness
Although the rule took effect immediately on the date of its publication occurred on January 15, 2025, the expected compliance date for these rules is set for May 15, 2025, with some elements of these controls having a delayed compliance date of January 15, 2026.
Comments