amd - Pwim.Net

Reuters: U.S. Aims to Thwart Chinese Access to AI Chips via Overseas Outlets

The Biden administration is contemplating measures to plug a loophole that has been enabling Chinese firms to acquire U.S. artificial intelligence (AI) chips through overseas channels, as reported by Reuters on October 13, 2023. This move underscores Washington’s ongoing effort to curb China’s burgeoning AI capabilities which are significantly anchored on U.S. chip technology.

Last year, the U.S. administration introduced restrictions on the shipment of AI chips and chip-making tools to China, aiming to impede its military advancements. The present consideration to broaden the scope of restrictions reveals the administration’s struggle to sever China’s access to top-notch AI technology amidst the intricacies in sealing every avenue in export controls.

In the initial phase of restrictions, the loophole left open permitted overseas subsidiaries of Chinese companies to have unrestricted access to these semiconductors, thereby, potentially enabling their smuggling into China or remote access by China-based individuals. Notably, chips forbidden by U.S. regulations were reportedly available from vendors in Shenzhen’s Huaqiangbei electronics area as of June.

The loophole’s existence points to the challenge faced by the U.S. in policing transactions involving these chips. While shipping the AI chips to mainland China is against U.S. law, enforcement becomes a hurdle as China-based personnel could lawfully access these chips housed at foreign subsidiaries.

Greg Allen from the Center for Strategic and International Studies highlighted that Chinese firms are procuring chips for overseas data centers, with Singapore emerging as a significant hub for cloud computing.

On the flip side, the Chinese government has previously voiced its discontent over U.S. export controls, accusing Washington of unwarranted suppression of Chinese enterprises.

China’s AI ascendancy is heavily reliant on its access to U.S. chips. A 2022 report by Georgetown University’s Center for Security and Emerging Technology revealed that a substantial portion of AI chips procured through Chinese military tenders were designed by U.S.-based giants like Nvidia, Xilinx, Intel, and Microsemi.

Furthermore, in August, the U.S. urged major chip manufacturers like Nvidia and AMD to limit the shipment of AI chips not only to China but to other regions including certain Middle Eastern countries. The upcoming rules, anticipated to be rolled out this month, are expected to extend these restrictions more broadly across all players in the market.

The Biden administration continues to wrestle with the task of closing loopholes, including the challenge of curtailing Chinese access to U.S. cloud service providers like Amazon Web Services (AWS), which extends similar AI capabilities.

In line with the ongoing U.S. efforts to curb Chinese access to advanced AI chips, as reported by Blockchain.News, senior Republican Representatives Michael McCaul and Mike Gallagher have urged for tighter export control enforcement. On October 6, 2023, they expressed concerns to the National Security Advisor over the Bureau of Industry and Security’s laxity in enforcing existing rules set in 2022 aimed at limiting China’s procurement of advanced semiconductors for military use. This call for stricter enforcement underscores the escalating U.S.-China tech tensions, particularly amidst advancements by China’s Semiconductor Manufacturing International Corporation and Huawei Technologies in semiconductor technology despite current U.S. sanctions.

US Toughens Export Controls on Semiconductors to China Amid Military Concerns

The US Department of Commerce’s Bureau of Industry and Security (BIS) unveiled tightened export controls on advanced computing semiconductors, semiconductor manufacturing equipment, and supercomputing items to nations deemed as “countries of concern”, including the People’s Republic of China (PRC) on October 17, 2023. This move aims to thwart the military modernization efforts pursued by the PRC, in a bid to shield national security interests. The latest raft of rules builds upon the initial restrictions enacted on October 7, 2022.

The BIS’s updated rules, effective November 16, 2023, augment the previously established controls by modifying the parameters that define a restricted advanced computing chip. Under the new guidelines, a chip will face export restrictions if it exceeds either of two delineated parameters: the performance threshold specified in the previous rule or a new “performance density threshold” introduced to preempt potential circumventions.

Additionally, the US government will assess certain chip exports falling just below the restricted threshold, following a mandatory notification by the exporters. This new framework also carves out an exemption, permitting the export of chips aimed at consumer applications.

A global licensing requirement for exporting controlled chips to any firm headquartered in a US arms embargoed country or Macau has been established to curb circumvention of the controls through foreign subsidiaries and branches. The updates also introduce new red flags and due diligence mandates for foundries to identify restricted chip designs emanating from countries of concern.

The licensing requisites extend to 22 nations under a US arms embargo and Macau, with a presumption of denial for advanced chips and a presumption of approval for other chips, bolstering the visibility for compliance monitoring and enforcement.

Parallelly, the BIS has expanded controls on different types of semiconductor manufacturing equipment and refined US persons restrictions to ensure US entities do not aid advanced PRC semiconductor manufacturing endeavors. The licensing requirements now encompass 21 other countries apart from the PRC and Macau.

On the same day, two PRC entities and their 13 subsidiaries, identified as being involved in the development of advanced computing chips detrimental to US national security and foreign policy interests, were added to the Entity List. Foundries manufacturing chips for these listed entities now require a BIS license.

Notably, the amplified restrictions will impact global semiconductor behemoths like Nvidia and AMD, whose certain chip models previously exempt, will now fall under the embargo. This is particularly significant given China’s position as a major market for these firms.

The rules are open for public commentary for 60 days post-publication, inviting insights on several facets including risks associated with Infrastructure as a Service (IaaS) providers, and additional compliance guidance for foundries.

Microsoft Reveals Azure Maia AI Accelerator and Azure Cobalt CPU

The Azure Maia 100 AI Accelerator and the Azure Cobalt CPU were both introduced at the Ignite 2023 conference, which was hosted by Microsoft. These two in-house silicon chips represent a significant advancement in artificial intelligence and cloud computing technologies. The Azure Maia AI Accelerator, which is specialized for AI and generative AI workloads, and the Azure Cobalt CPU, which is an ARM-based processor designed for general computing, represent a major step in Microsoft’s technological strategy. Both of these products are part of the Azure platform. These chips, which were created in the Redmond lab of Microsoft, are essential to the company’s goal of a completely integrated infrastructure, which combines software, servers, racks, and cooling systems.

These chips are slated for release in early 2024 and will be used to power various services provided by Microsoft such as Copilot and Azure OpenAI Service. Their incorporation into Microsoft’s data centers is a reaction to the increased need for computing capacity that is efficient, scalable, and environmentally friendly, particularly in cloud and artificial intelligence technologies.

Microsoft is adopting a comprehensive approach to infrastructure, with the goal of maximizing the performance of each individual component, from silicon to hardware to software. This approach is in keeping with the company’s aim of being more sustainable, one noteworthy example of which is the energy-efficient design of the Cobalt central processing unit (CPU).

Microsoft also announced an extension of its collaborations with NVIDIA and AMD, which would enhance the capabilities of both companies in the area of infrastructure and provide customers with a variety of alternatives in terms of both performance and cost.

Microsoft is working on developing second-generation versions of both the Azure Maia AI Accelerator and the Azure Cobalt CPU family in order to maintain its tradition of technological leadership in this industry. Their drive to expanding their skills in artificial intelligence and cloud computing is shown by their focus to optimizing each technological layer.

The recent advancements that Microsoft has made in the field of artificial intelligence chip technology were discussed in a related blog post. The Azure Maia AI Accelerator and the Azure Cobalt CPU, both of which are essential to the AI and cloud computing strategy developed by Microsoft, are expected to be operational somewhere in the early year 2024. This represents a larger trend in the artificial intelligence technology business, where major firms are increasingly focused on the manufacture of semiconductor chips for increased AI capabilities. This is shown by the fact that this is occurring.