Morgan Stanley: Only the GB200 AI server business generates $210 billion in annual revenue for NVIDIA

Morgan Stanley's latest analysis report estimates that the revenue from only the Blackwell "GB200" AI server orders will bring NVIDIA $210 billion in annual revenue.

The GB200 chip was released on March 19th, consisting of two B200 Blackwell GPUs and a Grace CPU based on Arm architecture. Its performance in large language models is 30 times higher than the H100, while costs and energy consumption are reduced to one twenty-fifth.

Morgan Stanley stated that the latest GB200 AI server orders are in short supply, and this situation is expected to continue beyond 2025.

NVIDIA's annual shipment volume of Blackwell "GB200" AI servers is between 60,000 to 70,000 units. Calculated at $2-3 million per server, the revenue from Blackwell servers alone could bring NVIDIA $210 billion in annual revenue.

NVIDIA's GB200 AI chip servers come in two specifications, NVL72 and NVL36. Due to more and more AI startups choosing the economically viable NVL36 servers, NVL36 is becoming more popular.

In June of this year, IT Home reported that the new GB200 series AI chips from NVIDIA are in short supply. After increasing the advanced process production volume with TSMC, NVIDIA also placed additional orders with backend testing factories. Orders from ASE Technology and King Yuan Electronics in the fourth quarter are expected to double compared to the previous quarter.

The testing process for the GB200 and B series AI chips is significantly longer than the previous H series, requiring four consecutive steps, including Final Test, Burn-in Aging Test, returning to Final Test, and finally undergoing SLT system-level testing