NVIDIA’s CEO Jensen Huang Delivers World’s First DGX H200 AI Station To OpenAI

NVIDIA has delivered the world's first DGX H200 AI station to OpenAI, being handed by none other than CEO Jensen Huang himself.

In an X post by Greg Brockman, President and co-founder of OpenAI, it was revealed that Team Green has finally handed over the firm's most powerful product to OpenAI. Interestingly, NVIDIA's CEO Jensen Huang arrived to deliver what we should call a technological marvel. OpenAI's CEO Sam Altman and Greg Brockman also received the NVIDIA package, showing that the DGX H200 is indeed something massive that requires high-profile attendees.

First @NVIDIA DGX H200 in the world, hand-delivered to OpenAI and dedicated by Jensen "to advance AI, computing, and humanity": pic.twitter.com/rEJu7OTNGT

— Greg Brockman (@gdb) April 24, 2024

Before going into what the DGX H200 offers, let's discuss industry dynamics. Apart from focusing on the fact that OpenAI did get exclusive access to the world's most potent AI system, it is interesting to witness how companies are battling against each other to get superior hardware out there. Recently, we talked a bit about Blackwell GPUs and how Meta placed the first orders of the next-gen architecture despite not being released to the mass markets. On the other hand, we have OpenAI, which focuses on building gigantic data centers with a proposed valuation of $7 trillion, but we won't go into that further.

The picture tells me two things; the first is that the future competition among firms isn't their AI revenue but the computing power they have onboard. The second is that the man in between, wearing the iconic leather jacket, will take the industry's spotlight for quite some time now because every company out there looks towards NVIDIA for their AI needs. We can't even debate it since the robust ecosystem Team Green provides to its customers is simply astonishing.

Coming back to the DGX H200, this beast of a machine is fueled by the prowess of HBM3E. It has memory capacities of up to 141 GB and up to 4.8 TB/s of bandwidth, which is 2.4x more bandwidth and double the capacity of the previous iteration. In a recent coverage, we highlighted the superiority of the H200s over the markets and how it managed to give almost double the performance in AI applications such as MLPerf v4.0, showing how dominant these pieces of hardware are in the market segment.

Interestingly, DGX H200 is based on the Hopper architecture which has been succeeded by the new Blackwell B100/B200 GPUs. These next-gen AI chips offer increased AI performance & capabilities & we are likely to see them in action later this year.