Oracle leads industry backlash, warning of dire consequences for US innovation, while others debate the effectiveness of AI regulation to address national security risks.
Credit: kropic1 / Shutterstock
The Biden administration’s recently introduced “Export Control Framework for Artificial Intelligence Diffusion” has sparked a heated debate in the tech industry. While the framework seeks to curb potential national security threats linked to AI and GPU exports, vendors including Oracle, Microsoft, Amazon, and Meta argue it imposes excessive regulatory burdens, stifling innovation and ceding market leadership to China.
Industry stakeholders believe that the framework may result in unintended consequences for the US and global tech ecosystems.
“It not only restricts China’s access to advanced technologies, pushing US firms to innovate and develop alternatives; it also limits US companies’ global market share and encourages China to accelerate its own technological advancements, altering the global tech landscape and intensifying the US-China tech competition,” said Charlie Dai, VP, principal analyst at Forrester.
The framework developed by the Bureau of Industry and Security (BIS), an agency of the US Department of Commerce, imposes strict licensing requirements on AI technology exports to address national security concerns, including the potential misuse of AI for developing weapons of mass destruction or enabling malicious activities.
A divisive regulatory proposal
The BIS’s new Interim Final Rule (IFR) mandates sweeping global export restrictions on AI technologies and GPUs, targeting potential misuse in high-risk areas such as weapons development and artificial general intelligence (AGI). However, industry leaders warn the framework may miss its mark.
The framework establishes a licensing system that restricts the number of GPUs US firms can export based on the total computing power deployed in recipient countries rather than individual use cases or quantities.
Under the rules, only a core group of 20 trusted countries, including most large European and Asian democracies, are exempt from export caps. However, key US partners including India, Singapore, Vietnam, and Mexico are excluded, along with smaller European nations such as the Baltic republics and Czechia, signaling disparities in US trust. Local cloud providers could qualify as “validated end users” by meeting specific security assurances, as seen in a recent US-UAE deal allowing G42 to access Microsoft’s advanced AI technology after limiting ties to China.
What it means for enterprises
The implications for enterprises are far-reaching. Under the new framework, businesses relying on GPUs for cloud services or AI development may face cost increases, supply chain challenges, and delays in accessing cutting-edge technologies. For cloud providers, compliance costs linked to retrofitting data centers with stringent security requirements could create additional burdens.
“For enterprises, the export control framework may disrupt GPU supply chains, causing project delays and increased operational costs, while forcing enterprises 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, risking market share loss as customers seek alternatives, and may experience innovation constraints due to the diversion of resources towards regulatory compliance,” he said.
美国总统拜登在离任前夕计划推出的一项AI芯片管制新规,在出台前先遭到了产业界的集体批评。
据环球时报援引彭博社1月9日的一篇报道,美国拜登政府计划在离任前出台一项新的人工智能芯片扩散出口管制框架规定,旨在对美国制造的AI GPU(图形处理器,主要用于AI大模型的训练及推理)芯片实施严格的全球出口限制。
这一框架将建立一个全球许可证制度,限制美国企业可以出口的AI GPU。
按照新规要求,美国将对各个国家及地区将根据其部署的芯片计算能力被划分为三个等级,不同等级适用不同的销售限制。
第一等级包括美国的主要盟友,如德国、荷兰、日本、韩国和中国台湾、新加坡、印度等18个国家和地区。这些国家几乎不受限制地使用美国厂商生产的AI芯片(注:美国厂商交给台积电等其他国家地区厂商代工生产的芯片也属于被管制范畴),并可以在其境内自由部署算力。
第二等级则包括除第一梯队外的绝大多数国家,这些国家将面临总算力限制,每个国家在2025年至2027年期间最多可获得约50000个AI GPU。
第三等级主要是中国大陆、俄罗斯及其他被美国实施武器禁运的国家及地区。这些国家将受到最严格的限制,几乎全面禁止进口美国厂商生产的AI GPU芯片。
此外,新规还有多项针对第三国转运的限制条款,防止第三等级国家通过不受限制的国家作“中转站”获得AI GPU。
据彭博社报道,由于特朗普将在1月20日正式宣誓就职,拜登在任期进入倒计时的最后几天尽力加速制定新规,计划最早于当地时间1月10日正式推出。
而在这项新规出台前,先招致了美国产业界的批评。
作为占全球AI GPU 90%以上市场份额的霸主,英伟达公开批评这项新规“毫无意义”“甚至对美国有害”。
在英伟达发给界面新闻的最新回应声明中,英伟达政府事务副总裁Ned Finkle表示,全球用户日常使用的个人电脑已经普遍搭载数据中心相关的GPU产品和技术,试图对其实施控制没有意义,反而有可能迫使全球转向寻找替代方案,损害美国利益。“我们鼓励拜登总统不要再此时制定一项只会损害美国经济的政策。”
另一家硅谷科技巨头甲骨文(Oracle)也“实名”反对新规。甲骨文执行副总裁Ken Glueck在公司官网上撰文称,这项新规可能成为美国科技行业历史上最具破坏性的政策之一。他直言不讳地表示,该政策“直接将美国公司的全球芯片市场缩小了80%”,等于“将芯片市场拱手让给中国等竞争对手”。
当前,英伟达、甲骨文等硅谷科技巨头都在全球范围内积极建设AI数据中心。
英伟达CEO黄仁勋过去一年先后拜会印度、越南、泰国、新加坡等国家政要,签署了多项支持各国建设本土AI数据中心的协议。按照计划,英伟达将在未来五年内投资3万亿美元,用以建设全球数据中心。
据公司官方统计,甲骨文在全球运营及在建的云数据中心数量已超过160个,这些数据中心配备了大量的英伟达AI GPU集群,用于训练大型AI模型。
早从2023年10月开始,美国政府已出台了多轮针对中国大陆出口AI芯片的管制规定,英伟达的A100和H100两款GPU及后续推出的降配版(A800与H800)均被限制出口,AMD、英特尔旗下同类的AI芯片产品也都被纳入其中。
美国半导体行业协会(SIA)此次也选择公开发声,表达了对这项新规的担忧。SIA发布公告称,这一政策的潜在监管行动将对美国先进集成电路的出口带来全球限制和繁重的许可要求。机构同时指出,该法规的范围和复杂性是“前所未有的”,在没有与行业进行充分磋商的情况下,拜登政府在即将卸任的最后关头“不该仓促做出如此重大的政策转变”。
SIA呼吁政府应该发布一项拟议规则,或将政策制定过程移交给即将上任的特朗普政府,以确保有适当机会进行深入讨论。
对于美国多次升级的芯片出口管制政策,中国此前已多次表明抗议立场。
去年12月,由于美国不断升级对华芯片出口管制,中国半导体行业协会、中国互联网协会、中国半导体行业协会、中国汽车工业协会和中国通信企业协会齐发公告,称“美国芯片不再安全,不再可靠”,呼吁中企审慎采购美国芯片。
商务部新闻发言人何亚东对此评价表示,美方不断泛化国家安全概念,滥用出口管制措施,实施单边霸凌行径,严重破坏市场规则和国际经贸秩序,对半导体领域正常贸易往来带来极大不确定性,对全球产业链供应链稳定构成了威胁。中方将采取必要措施,坚决维护自身正当权益。