The United States further restricts the export of high-performance computing GPUs

Last year, NVIDIA, in a bid to circumvent export restrictions, unveiled its A800 series of computation cards, tailored exclusively for the Chinese market. When juxtaposed with the extant A100 series, the specifications of the A800 series largely mirror their predecessors, with the most salient disparity being the connection speed of the NVLink interconnect bus: the former boasts 600 GB/s, while the latter is curtailed at 400 GB/s. Subsequently, this year saw the advent of the H800 series, crafted to satiate export control stipulations.

Recent media revelations disclose that NVIDIA, in a regulatory document released this week, acknowledged receiving official communiques during the second quarter of the 2024 fiscal year. These directives stipulate augmented licensing prerequisites for the sale of A100 and H100 computation cards to certain regions, encompassing the Middle East. NVIDIA reaffirmed the unfettered availability of alternative offerings in China, including the A800 and H800 cards.

Although NVIDIA elucidated these new impositions in the regulatory dossier, it abstained from enumerating the specific nations or territories to be impacted. In recent years, Saudi Arabia and the UAE have been fervently augmenting their prowess in artificial intelligence, acquiring NVIDIA’s computational cards in significant quantities, thus emerging as pivotal clientele. These new edicts will inevitably resonate with them. Of paramount import is the underlying objective: to forestall the possibility of A100 and H100 cards, destined for Middle Eastern locales, from ultimately permeating the Chinese market.

NVIDIA postulates that these nascent restrictive measures won’t inflict a substantial dent in its revenue stream and is currently exploring collaborative avenues to address these associated quandaries.