On April 5, 2016, NVIDIA announced the DGX-1, which it billed as “the world’s first deep learning supercomputer.” The system packed eight Tesla P100 GPU accelerators into a single box delivering up to 170 teraflops of half-precision (FP16) performance, and it was designed to “dramatically reduce the time to train larger, more sophisticated deep neural networks.” Instead of stitching together loose GPUs, the DGX-1 was a turnkey AI machine that data scientists could plug in and use.
The first DGX-1 went to OpenAI, the research lab founded only months earlier in late 2015. Jensen Huang personally hand-delivered the unit to OpenAI’s San Francisco office, an episode NVIDIA has since framed as the moment that “ignited the modern AI revolution.” OpenAI used the machine to accelerate research experiments that would otherwise have taken far longer on conventional hardware.
The DGX line became NVIDIA’s reference design for AI infrastructure and the seed of a business that now sells entire data-center systems rather than individual chips. The 2016 hand-delivery also marked the start of a relationship between NVIDIA and OpenAI that would grow into some of the largest compute deals in the industry.
Why business readers should care: the DGX-1 shows a hardware vendor moving up the value chain from selling components to selling complete systems - and seeding a strategic customer at its earliest, most fragile stage. Backing the right lab before it was famous turned into enormous leverage as that lab’s compute appetite exploded.