NVIDIA announced the DGX-1, which it billed as “the world’s first deep learning supercomputer,” on April 5, 2016. The system held eight Tesla P100 GPUs delivering up to 170 teraflops of FP16 performance in a single box. The very first unit went to OpenAI, the research lab founded only months earlier, and NVIDIA CEO Jensen Huang hand-delivered it in person to the lab’s San Francisco office.
NVIDIA has repeatedly highlighted the hand-delivery as a symbolic origin point for the modern AI era - the chipmaker placing its first turnkey AI machine directly into the hands of the lab that would go on to build ChatGPT. It also seeded a relationship between NVIDIA and OpenAI that grew into some of the largest compute commitments in the industry.