NVIDIA launched CUDA, enabling general-purpose GPU programming; without it, modern deep learning would have been impractical.
In 2006–2007, NVIDIA launched CUDA, a programming model for general-purpose GPUs. ML researchers soon found that training large nets fit GPU SIMD parallelism.
Later frameworks (Theano, cuDNN, TensorFlow, PyTorch) rode that stack. Cost per FLOP fell and model size grew.
Today chip geopolitics and NVIDIA’s roadmap are as much part of AI history as any architecture paper.
Per an internal memo reviewed by Reuters in July 2026, Meta planned to begin manufacturing its Iris accelerator from September within a four-generation MTIA program, with Broadcom on design and TSMC on fabrication. September coverage indicates production has started; infrastructure ambition is ~7 GW in 2026 and double (14 GW) in 2027, supplementing—not replacing—Nvidia and AMD GPUs.
On July 24, 2026, Jensen Huang used his first X post to share the “Open Weights and American AI Leadership” letter, signed by Nvidia, Microsoft, Meta, Mistral, Hugging Face, the Linux Foundation and about 25 organizations total; OpenAI, Anthropic, Google and Amazon were absent from the original list. In parallel, Huang confirmed Vera Rubin was already in production, with shipments expected in the fall.
On 2 October 2026 (press spotlighted on the 5th), ASUS announced the 64GB Ascent GX10 configuration on the NVIDIA DGX Spark platform with the GB10 Grace Blackwell Superchip, up to 1 petaFLOP, coherent unified memory for 30–35B open-source models, and two-unit clustering toward 128GB and roughly 1.7× throughput.