AI is now also transforming the manufacturing of the very chips on which it runs. At the SEMICON West trade show in San Francisco on October 14, NVIDIA demonstrated how chip manufacturers and their equipment suppliers are using AI from development through testing. The company refers to “AI-defined fabs,” meaning semiconductor plants where AI supports engineers in process optimization, troubleshooting, and improving efficiency. Applied Materials, Cadence, KLA, Lam Research, Siemens, and Synopsys are working with NVIDIA’s software libraries for this purpose. A new addition is cuPotential, which enables AI agents to simulate how atoms and molecules interact with one another. In early-access testing, the library has already accelerated GPU-optimized workflows by a factor of two to three.

“AI is now changing how chips are built”

The remaining figures come from the participating companies. Lam Research expects to compute selected wafer etching models up to 30 times faster; simulations for plasma predictions there now take only milliseconds instead of days in some cases when using NVIDIA software. Applied Materials reports chamber simulations that are up to 35 times faster and optimizes layouts and material flow in digital twins of its factories. In collaboration with KLA, an image-processing kernel for inspection was developed that operates 50 times faster than earlier CPU versions. Teradyne detects worn test probes about 85 times faster before they reduce yield. “The semiconductor industry made the AI revolution possible. Now AI is changing how chips are built,” says Tim Costa, Vice President of Industrial and Computational Engineering at NVIDIA.

Robotics and quality inspection are also part of this. FANUC and Foxconn test robotic workflows in digital factory twins on NVIDIA’s Omniverse platform before putting them into operation. OMRON uses world models from NVIDIA’s Cosmos family of agents in its inspection devices to assist engineers with troubleshooting printed circuit boards.

Test Sites in China

Siemens is making the transition from simulation to live production in China. At the China International Industry Fair in Shanghai, the company unveiled six application scenarios for Physical AI across five plants. AI-driven systems are being tested there under real-world conditions, including the sorting of unordered bulk materials, autonomous container transport, palletizing, quality inspection, precision screw fastening, and flexible packaging. Together with partners, Siemens aims to investigate which capabilities can be transferred to other production lines and operating environments. In addition, the company introduced more than 20 new products—13 of which were developed in China—signed cooperation agreements with over 40 Chinese companies, and unveiled a new version of its proprietary Engineering Agent, which, according to Siemens, can reduce program development time by up to 50 percent. Together with a Chinese mechanical engineering research institute, Siemens also published a national standard for reconfigurable factory systems. It provides plants, production lines, and software with a common language for their capabilities and is intended to facilitate the use of digital twins and AI models.

Rearrangement in Minutes

The San Francisco-based startup Industrial Next aims to demonstrate just how much faster manufacturing could be adapted in the future.

The founders, Lukas Pankau and Allen Pan, previously worked at Tesla. They are building self-contained robotic stations controlled by their own foundation models for image recognition, task understanding, and movement. The company promises that a production line can be retooled in minutes instead of weeks—without traditional programming. At the TechCrunch Disrupt conference, the company announced a collaboration with the BMW Group Technology Office, which is testing the stations’ fine motor skills for manufacturing.

Practice Before Strategy

The Industrial Index 2026 from software provider IFS—which surveyed more than 2,200 decision-makers in 14 countries—examines how far practice is ahead of strategy. Over 90 percent of companies report significant improvements in reliability and performance, such as in equipment availability and on-time delivery. 43 percent of respondents have trained AI agents and integrated them into their daily work over the past twelve months, driven primarily by operational teams. During this period, senior management focused primarily on building their own AI expertise. The study reveals gaps in security measures. More than 70 percent of companies express confidence in their security and data policies, yet only 11 percent have validated AI results for security- or quality-critical processes. Only 4 percent require human approval for AI-supported decisions. Since IFS itself sells industrial AI software, this is a vendor-sponsored study. Nevertheless, its findings align with other reports from this week. AI has arrived on the factory floor; governance and control must now follow suit.