Report: Chinese organizations use public cloud to access restricted AI chips – SiliconANGLE

Has the hype around generative AI peaked? Learn how alternatives like agentic AI are gaining traction in the evolving AI landscape.



Multiple organizations in China have sought to rent high-end graphics cards from U.S. cloud providers, Reuters reported late Thursday.

The news agency learned of the procurement effort by reviewing publicly-available tender documents. Those are invitations that ask companies to submit bids for a contract, in this case artificial intelligence infrastructure purchases.

Under U.S. export controls that rolled out two years ago, Nvidia Corp. is prohibited from selling its high-end H100 and A100 graphics processing units to entitles in China. The rules also cover certain other data center chips optimized for AI workloads. But though such chips can’t be sold or transferred to entities based in China, leasing them via the cloud is not prohibited. 

The documents reviewed by Reuters indicate that at least 11 Chinese organizations have sought access to “restricted U.S. technologies or cloud services.” Many of those organizations planned to complete the transactions through intermediaries.

According to Reuters, a university spent the equivalent of $28,000 to access Amazon Web Services Inc. instances equipped with A100 and H100 chips. In April, a research institute developing a custom large language model disclosed plans to spend a similar sum on AWS infrastructure. It intended to use the hardware to power an AI model.

Microsoft Corp.’s competing Azure cloud was also named in the public tender documents. In one of the filings, a university stated that it had purchased access to OpenAI models hosted on Azure. The deal covered 40 billion tokens’ worth of AI processing capacity.

Like Microsoft, AWS offers access to third-party AI models in its cloud. According to Reuters, the Amazon.com Inc. unit published a number of Chinese-language marketing documents that highlighted an opportunity to try out “world-class AI models” on its platform. The news agency reported that some of the marketing materials, which included blog posts, referenced Anthropic PBC’s Claude series of LLMs. 

Following the report, AWS updated dozens of blog posts with a note stating that not all its services are available via its cloud regions in China. The Amazon unit told Reuters that “Amazon Bedrock customers are subject to Anthropic’s end user license agreement, which prohibits access to Claude in China both via Amazon’s Bedrock API (application programming interface) and via Anthropic’s own API.”

An AWS spokesperson said that “AWS complies with all applicable US laws, including trade laws, regarding the provision of AWS services inside and outside of China.”

The report comes as U.S. policymakers take steps to more closely regulate cloud-hosted AI models and chips.

In January, the Commerce Department proposed a rule that would require cloud providers to be more transparent about how customers use their platforms to train AI models. The regulation would place a particular emphasis on AI models capable of “malicious cyber-enabled activity.” More recently, lawmakers submitted a bill that would enable regulators to more closely scrutinize how technologies such as AI chips are accessed via the cloud. 

Photo: Unsplash

Your vote of support is important to us and it helps us keep the content FREE.

One click below supports our mission to provide free, deep, and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy

THANK YOU



Source link
lol

By stp2y

Leave a Reply

Your email address will not be published. Required fields are marked *

No widgets found. Go to Widget page and add the widget in Offcanvas Sidebar Widget Area.