The first person convicted of AI espionage wasn’t stealing code. He was stealing the designs for how the code gets trained.
Last week, a federal jury in San Francisco convicted Linwei Ding, a former Google engineer, on fourteen counts of economic espionage and trade secret theft.
Over nearly a year, he copied more than 2,000 pages of designs for Google’s TPU chips, GPU systems, and networking hardware to his personal cloud account.
The architecture that trains AI. Not the models themselves.
While still at Google, Ding secretly founded an AI company in Shanghai and applying to a government-backed talent programme. He told Chinese investors he could replicate Google’s supercomputing infrastructure.
The FBI called it the first conviction on AI-related economic espionage charges. It won’t be the last.
What makes this case worth watching isn’t the crime. Industrial espionage is as old as industry. It’s what was worth stealing.
Not a model, which depreciates in months. Not code, which gets rewritten.
The training infrastructure. The thing that is hardest to build, hardest to replace, and currently at the centre of a global spending race worth hundreds of billions.
This isn’t about code. It’s about the factory where the code is born.
— ♻️ Repost for anyone who missed the shift from software to infrastructure.
