Nvidia has released data showing its advanced server system delivers 10 times better performance for open-source “mixture-of-experts” AI models from DeepSeek, Mistral and Moonshot AI.
American chip giant Nvidia (NVDA) on Wednesday released data showing that its GB200 NVL72 advanced server system delivers 10-fold performance gains to open‑source AI models from China’s DeepSeek AI and Moonshot AI as well as France’s Mistral AI.
The NVL72 is a setup where 72 Blackwell‑generation graphics processing units (GPUs) are linked so they operate as a single AI supercomputer within one rack, combining hardware and software optimizations.
Nvidia said this system delivers around 10 times the performance of its prior‑generation H200‑based platform for DeepSeek’s DeepSeek‑R1, an open‑source “mixture‑of‑experts” (MoE) AI model that routes parts of each query to different specialized sub‑models to handle complex tasks.
Moonshot AI’s Kimi K2 Thinking and Mistral AI’s Mistral Large 3 — similarly considered as top open-source MoE models — also achieved roughly 10‑fold performance improvements on GB200 NVL72 compared with deployments on H200‑class Hopper systems.
Google to Rival Nvidia with Custom TPU
The release of the data comes at a time Alphabet’s (GOOGL) Tensor Processing Units has emerged as a rival to Nvida’s GPUs. Alphabet’s Google is reportedly in talks with Meta (META) for take-up of its TPUs. The TPU is Google’s in-house chip for large AI workloads.
However, responding to the move, Nvidia’s Newsroom account on X noted that “Nvidia is a generation ahead of the industry”. The post touted that Nvidia is “the only platform that runs every AI model and does it everywhere computing is done.”
“NVIDIA offers greater performance, versatility, and fungibility than application specific integrated circuits (ASICs), which are designed for specific AI frameworks or functions,” Nvidia said.
Meanwhile, Nvidia this week also launched what it called “the world’s first industry-scale” open-source software for the development of self-driving vehicles.
Is NVDA Stock a Strong Buy?
On Wall Street, Nvidia’s shares remain a Strong Buy with 39 Buys, one Hold, and one Sell assigned by 41 analysts over the past three months.
At $258.10, the average NVDA price target indicates approximately 43% growth potential from the current trading levels.
