SAN FRANCISCO, Oct. 14, 2026 | NVIDIA said Wednesday that semiconductor companies are adopting its CUDA-X libraries to build what it calls “AI-defined fabs,” factories where AI helps engineers refine processes, fix production problems and raise output. The announcement, made during SEMICON West, centres on a new library called cuPotential, according to a NVIDIA press release published on Yahoo Finance Canada. The performance figures come from NVIDIA and its partners, and the release offers no independent verification.

cuPotential is in early access and lets developers build AI agents that simulate how atoms and molecules interact. NVIDIA reports an additional 2x to 3x speedup on workflows that were already GPU-optimized, based on early testing. Other tools in the release include NanoVDB, an open-source library for 3D data that is available now, and NPP, an image-processing library available in the CUDA Toolkit.

Tim Costa, NVIDIA’s vice president and general manager of industrial and computational engineering, said: “The semiconductor industry made the AI revolution possible. Now, AI is transforming how chips are built.”

NVIDIA lists a long roster of partners and results, all reported by the companies themselves. Synopsys, an early adopter of cuPotential, reports a potential 30x speedup with NVIDIA’s cuEST library on its QuantumATK software and 12x with cuDSS on S-Device benchmarks. Cadence says its AuraStack AI Super Agent optimizes advanced packages 8x faster using cuDSS, and Applied Materials reports up to 35x faster chamber simulations and fab digital twins. The research institute imec reports a 30% improvement in its lithography process window using NVIDIA’s cuLitho library and AI-based image noise removal.

Lam Research plans to add NanoVDB to its SEMulator3D platform, with up to 30x faster wafer-etch modelling, targeted for release by the end of 2026. The speedups apply to specific workloads and benchmarks, not to overall fab performance, and many are labelled “potential” or come from early-access testing. The release also carries standard forward-looking-statement warnings, noting that actual results may differ materially.

What is an AI-defined fab?
It is NVIDIA’s term for a semiconductor factory where AI helps engineers refine processes, fix production problems and raise output, using NVIDIA’s CUDA-X libraries.

Are the speedup figures independently verified?
No. The figures come from NVIDIA’s press release and are reported by NVIDIA and its partners, many from early-access tests or benchmarks of specific workloads.