Former Google software engineer Linwei Ding has been convicted by a US court of stealing confidential information related to the company’s artificial intelligence infrastructure. Prosecutors accused Ding of taking sensitive AI-related secrets from Google and sharing them with Chinese technology companies. He has been sentenced to 12 months minus one day in prison.
The information Ding transferred included details about Google’s hardware infrastructure and software systems used to train large AI models, including its Tensor Processing Units (TPUs) and Graphics Processing Units (GPUs). Google’s custom-designed TPUs are a key part of its AI infrastructure, enabling the company to train models without depending entirely on outside computing providers.
Experts have said that access to such documents could allow another company to bypass significant portions of the development process for an AI supercomputer that may have taken Google years and millions of dollars to build.
US District Judge Vince Chhabria said the sentence reflected the seriousness of the offence and described Ding’s actions as a systematic and blatant attempt to steal Google’s intellectual property. In addition to his prison sentence, Ding will serve two years of supervised release and has been ordered to pay more than $189,000, or around Rs 1.79 crore, in restitution along with a $25,000 fine, approximately Rs 23.75 lakh.
Who is Linwei Ding?
Linwei Ding, also known as Leon Ding, is a Chinese national who moved to the US in 2010 to pursue a master’s degree. He subsequently worked at several technology companies in Silicon Valley before joining Google as a software engineer in 2019. He became a lawful permanent resident of the US in 2021.
Prosecutors said his position at Google gave him access to sensitive information concerning the hardware and software infrastructure powering the company’s supercomputing data centres, which are used to train large AI models.
A San Francisco jury found Ding guilty in January following an 11-day trial. He was convicted on seven counts of trade secret theft and seven counts of economic espionage, although the espionage convictions were later set aside by the judge.
How did Linwei Ding steal Google’s AI secrets?
Prosecutors alleged that Ding began moving confidential information from Google’s internal network to a personal Google Cloud account on May 21, 2022, continuing the activity until May 2, 2023. During the trial, prosecutors said he uploaded more than 500 confidential files over roughly a year.
A later superseding indictment alleged that the material consisted of more than 1,000 separate files, while prosecutors presented evidence involving over 2,000 pages of confidential information. Court documents further stated that in December 2023, less than two weeks before leaving Google, Ding downloaded the information onto his personal computer.
The stolen files reportedly contained information about Google’s infrastructure and software for training large AI models, including technical details concerning TPUs and GPUs. They also included software that enables chips to communicate and perform tasks, as well as systems designed to coordinate thousands of chips into a supercomputer.
Other information allegedly involved Google’s custom SmartNIC, a networking component used in its AI supercomputers and cloud networking products. Prosecutors argued that some of the stolen material was intended to help Google maintain an advantage over competing cloud companies while reducing its dependence on Nvidia chips.
Ding’s connections with Chinese companies while he was still employed at Google were also central to the case. Prosecutors said that several months after he began uploading the confidential files, a Chinese start-up offered him a position as chief technology officer. Ding visited the company, Rongshu, in late 2022 and participated in investor meetings from around December.
In 2023, he established Zhisuan, a China-based technology company that proposed developing and training large AI models. According to the indictment, a Zhisuan document circulated by Ding in November 2023 stated that the company had experience with Google’s large-scale computing platform and planned to replicate, upgrade and adapt it for China.
Ding also allegedly applied in late 2023 to a Chinese government programme designed to attract technology talent to the country. Prosecutors claimed that he did not disclose his relationships with either company to Google. While in China promoting Zhisuan, he allegedly arranged for an intern to swipe his Google access card multiple times so that it appeared he was working from his office.
However, the judge later concluded that the evidence did not establish that Ding knew or intended during the May 2022 to April 2023 period that his actions would benefit the Chinese government.
The case comes amid an intensifying AI technology race between the US and China. The US has imposed restrictions on China’s access to some of the most advanced GPUs used for AI training, while Chinese technology companies are increasingly working to develop domestic alternatives.
A recent AI model called Ox Alpha, which gained attention online, was reportedly trained entirely using chips manufactured in China. The model is part of Z.ai’s GLM-5 family.
Meanwhile, an account associated with Chinese state-run media recently criticised US influence over AI development ahead of joint talks between the two countries, arguing that American technology companies should not be able to set AI rules for the rest of the world.
