Nvidia advances AI software diversification and AI‑assisted chip design

Developers are using the Vulkan graphics API and the Mojo programming language to run AI models on a wide range of GPUs—including AMD, Intel and Apple Silicon—challenging Nvidia's long‑standing CUDA monopoly. Vulkan enables cross‑vendor execution in projects such as llama.cpp, while Mojo, created by Chris Lattner's company Modular, lets the same code target both CUDA and other GPU stacks. Qualcomm’s near‑$4 billion acquisition of Modular underscores the commercial interest in these technologies.

At the same time, Nvidia is deploying AI agents to accelerate its own chip‑engineering workflow. Vice president Tim Costa highlighted that future chip complexity will outpace traditional design methods, prompting the use of AI for simulation, verification and optimization across architecture, packaging and system integration. Nvidia sees AI as essential infrastructure for delivering the next generation of high‑performance silicon.