中国获取人工智能算力的能力明年将翻倍，但其中约65%将来自美国设计的芯片，这表明中国仍持续依赖外国硬件，一项新研究显示。
China’s access to artificial intelligence computing power will double next year, but US-designed chips will account for about 65 per cent of that capacity, an indication of its continued reliance on foreign hardware, according to new research.
总部位于华盛顿的进步研究所（IFP）的一份报告称，预计2027年中国实体可用的总算力将从约460万扩大至880万“H100等效值”——这一指标以英伟达的数据中心图形处理单元为基准。
Total compute available to Chinese entities was expected to expand in 2027 from about 4.6 million to 8.8 million “H100-equivalents” – a metric benchmarked against Nvidia’s data-centre graphics processing units, according to a report by the Washington-based Institute for Progress (IFP).
为预测这些数字，进步研究所表示，其统计了国产芯片、合法进口、走私处理器以及通过海外云网络远程访问的算力。根据本周发布的报告，仅远程访问预计到2027年底将增至280万H100等效值，高于今年的约100万。
To project these figures, IFP said it tallied domestic chips, legal imports, smuggled processors and compute accessed remotely via overseas cloud networks. Remote access alone was expected to swell to 2.8 million H100-equivalents by the end of 2027, up from roughly 1 million this year, according to the report published this week.
报告称，尽管美国在AI训练算力上保持着约16倍的优势，但中国领先模型落后美国竞争对手仅6.5个月。
While the US maintains a roughly 16-fold advantage in AI training compute over China, leading Chinese models lag their American rivals by only 6.5 months, the report said.
进步研究所指出，蒸馏技术——即开发者利用优秀AI模型的输出训练系统——是帮助中国缩小差距的关键因素。中国此前曾拒绝美国关于国内开发者进行“激进、恶意”AI蒸馏的指控，称其毫无根据。
IFP pointed to distillation – where developers train systems on the outputs of superior AI models – as a key factor helping China close the gap. China has previously rejected US allegations of “aggressive, malicious” AI distillation by domestic developers as groundless.
鉴于这些发现，该研究所敦促美国决策者对海外云访问、芯片走私和合法出口实施更严格的管控。
In light of the findings, the institute urged US policymakers to impose stricter controls on overseas cloud access, chip smuggling and legal exports.
华盛顿政策组织“AI政策与战略研究所”周二发布的一份单独报告指出，美国设计的算力仍可合法租赁给原本被禁止购买该硬件的外国实体。
A separate report published on Tuesday by the Institute for AI Policy and Strategy stated that US-designed compute power remained legally available for rent to foreign entities otherwise barred from buying the hardware.
该政策组织呼吁美国工业与安全局限制中国获取足以训练前沿AI模型或大规模运行此类模型的云算力，同时主张加强对客户身份和所有权的核查。
The Washington-based policy group called on the US Bureau of Industry and Security to restrict Chinese access to cloud capacity large enough for training frontier AI models or running them at scale, while advocating for stronger checks on customer identity and ownership.
虽然芯片出口受到许可限制，但华盛顿在1月份采取行动，允许在安全条件下对英伟达的H200、超威半导体的MI325X以及同类处理器进行逐案出口审查。
While chip exports are subject to licensing restrictions, Washington moved in January to allow case-by-case export reviews for Nvidia’s H200, Advanced Micro Devices’ MI325X and comparable processors under security conditions.
美国出口商必须证明销售不会导致供应从美国客户处转移，而中国买方必须采用正式的出口合规程序，并将芯片提交在美国境内进行独立测试。
American exporters must show that sales would not divert supply from US customers, while Chinese buyers must adopt formal export compliance procedures and submit chips for independent testing within the US.
与此同时，中国科技巨头正争相加强国产替代。华为技术上个月表示，其专为AI模型训练设计的升腾960DT芯片将于2027年第一季度上市——比原定计划提前了三个季度。
Meanwhile, Chinese tech giants are scrambling to bolster domestic alternatives. Huawei Technologies last month said its Ascend 960DT chip, designed for AI model training, would become available in the first quarter of 2027 – three quarters ahead of its original schedule.
这家总部位于深圳的公司透露，其Atlas 900 A3 SuperPoDs已部署超过1000套，而Atlas 950系统已进入大规模商用。这些集群架构将数千个单独的处理器连接起来，以处理繁重的AI工作负载。然而，扩大国产硬件生产仍面临挑战。
The Shenzhen-based firm revealed that more than 1,000 of its Atlas 900 A3 SuperPoDs had already been deployed, while its Atlas 950 system had entered large-scale commercial use. These cluster architectures link thousands of individual processors to handle demanding AI workloads. However, scaling domestic hardware production remains a challenge.
总部位于旧金山的研究机构Epoch AI在上个月更新的报告中指出，高带宽内存——向AI芯片供数据的关键组件——是制约华为产能的主要瓶颈。
In an updated report last month, San Francisco-based research group Epoch AI identified high-bandwidth memory – a crucial component for feeding data to AI chips – as the main constraint on Huawei’s output.
Epoch AI估计，华为今年的芯片产出所提供的算力将不足英伟达全球出货量算力的4%。
Epoch AI estimated that Huawei’s chip output this year would deliver less than 4 per cent of the computing power of Nvidia’s global shipments.