AI Espionage: How Ex-Google Engineer Stole Supercomputer Blueprints

A federal jury in San Francisco has convicted a former Google software engineer on multiple counts of economic espionage and trade secret theft, marking the United States' first-ever AI-related espionage conviction. Linwei Ding, also known as Leon Ding, 38, was found guilty on seven counts of economic espionage and seven counts of theft of trade secrets. Court documents reveal that Ding stole thousands of pages of confidential AI technology from Google, allegedly to benefit the People's Republic of China.

According to the Department of Justice, the stolen trade secrets included in-depth details of Google's custom Tensor Processing Unit (TPU) architecture and its graphics processing unit (GPU) systems. They also contained information on Google's proprietary SmartNIC, a specialised network interface card that facilitates high-speed communication across the company's AI supercomputers and cloud networks.

Aakash Gupta, a product growth expert and former Vice President of Product at a Unicorn company, highlighted the significance of the AI espionage in a post on X. He wrote that Ding stole the “same infrastructure Anthropic just committed $42 billion to access,” and added, “The same technology analysts now value at $900 billion as a standalone business.”

Gupta highlighted that the case illustrates the real bottleneck in AI advancement. “China can manufacture chips. China can train models. What China cannot easily replicate is the 15 years of institutional knowledge Google accumulated about chip-to-chip communication, thermal management, and software orchestration across 9,000+ chip clusters,” he wrote.

Ding allegedly misrepresented his expertise to Chinese investors, claiming he was “one of 10 people in the world” capable of replicating Google's AI supercomputing platform. Gupta said, “The strategic value of TPU architecture isn't the silicon. It's the system-level engineering that turns silicon into intelligence.”

The case shows that AI espionage is about more than stealing documents or intellectual property. It involves acquiring the human capital and know-how that enables breakthrough technologies. “You can steal blueprints, but you can't steal a decade of build-test-iterate cycles,” Gupta wrote.

He concluded his post by saying, “Export controls block chip shipments. This case blocks something harder to monitor: the human capital that makes those chips useful.”

Here's Gupta's post:

The first-ever AI espionage conviction tells you exactly how the AI race will be fought.

Linwei Ding stole 2,000 pages of Google TPU architecture while having an intern badge-swipe him into work from 6,000 miles away. The jury took three hours to convict on all 14 counts. He… https://t.co/GrvHyOPSuC

— Aakash Gupta (@aakashgupta) January 31, 2026

Ding, who is scheduled to appear in court on February 3, faces a potential maximum sentence of 10 years for each count of trade secret theft and 15 years for each count of economic espionage, according to the Department of Justice.