Nvidia settled a case that started when a former hire walked into the company with autonomous-driving trade secrets he had taken from his old job at Valeo. The deal became public in a filing in federal court in San Jose, where both sides told the judge they had resolved the fight. The court then canceled the trial that had been scheduled for next month. No terms were released, and neither side has said what they agreed to, but the dispute came from a joint project for Mercedes-Benz, where both companies were working on tech for advanced driving features. In 2021, Valeo said one of its own engineers left for Nvidia and later joined internal work at the chip company. During a video call, Valeo staff spotted their verbatim source code on his screen, and they took a screenshot before he closed the window. That moment became the core of a lawsuit Valeo filed in 2023, accusing Nvidia of benefiting from material that never should have left its servers. Judge sends case forward after reviewing evidence A filing from Nvidia rejected the claim that it used any stolen code to build its parking-assist system. The company said it “rolled back” every task the engineer, Mohammad Moniruzzaman, touched. Nvidia said it also cut ties with Mohammad once the issue was confirmed. Mohammad was later convicted in Germany for infringing business secrets tied to Valeo’s software. A judge reviewed the discovery record and said there was enough “circumstantial” evidence to let a jury hear the case. That ruling landed in August and gave Valeo a path to argue that Nvidia gained value from the confidential files. The case sat under the title Valeo Schalter und Sensoren GmbH v. Nvidia Corp., 23-cv-05721, at the US District Court for the Northern District of California. Google pushes TPU support to challenge Nvidia While Nvidia was closing the legal chapter with Valeo, Google is coming for its throne, rampantly building a new push inside the company called TorchTPU to make its chips better at running PyTorch, the world’s most common AI software framework. Alphabet CEO Sundar Pichai told shareholders at the Q3 event that his goal is to remove the barriers that made developers stick to Nvidia hardware. Google wants its Tensor Processing Units to serve as a real alternative to Nvidia GPUs, which still dominate data-center installs for machine-learning work. TPU sales feed Google’s cloud revenue, and the company wants investors to see returns from its AI budget. TorchTPU aims to make TPUs fully compatible with the tools developers already use. Some teams inside Google are also debating whether to open-source parts of the software to speed adoption. PyTorch, backed heavily by Meta, sits at the center of modern AI development. In Silicon Valley, few engineers write low-level instructions for chips from Nvidia, AMD, or Google. Instead, they rely on frameworks with ready-made code. PyTorch launched in 2016 and grew alongside CUDA, the software stack many analysts say shields Nvidia from rivals. Nvidia’s teams have spent years making PyTorch run smoothly on its chips. Google, by comparison, trained its engineers on Jax, paired with a tool called XLA for performance on TPUs. That internal focus created distance between how Google builds AI systems and how customers actually write their models. TorchTPU is meant to close that gap and give companies a reason to shift workloads off Nvidia hardware. Sign up to Bybit and start trading with $30,050 in welcome gifts
