Microsoft's demand for NVIDIA GB200 AI servers is more than all other cloud companies combined

Microsoft's demand for NVIDIA's new GB200 AI servers "exceeds total orders from other CSPs" or cloud service providers, with the Windows' giant Q4 2024 orders "significantly" increased 3-4x says insider Ming-Chi Kuo.

In a new post on Medium, the insider points out that Microsoft's GB200 key component suppliers will begin mass production and shipments starting in Q4 2024, contributing to the supply chain's performance earlier than competing CSPs.

Kuo expects around 150,000 to 200,000 x Blackwell AI GPU chip shipments in Q4 2024, with "significant growth projected" at 200-250% quarter-over-quarter to 500,000 to 550,000 units in Q1 2025.

Microsoft is currently the "most aggressive customer" in procuring GB200, with Q4 2024 orders for GB200 surging 3-4x from the previous 300-500 racks (mostly NVL36) to around 1400-1500 racks (70% of them NVL72) with subsequent orders from Microsoft to "primarily focus" on NVL72 AI server cabinets.

Kuo says that Microsoft will initially deploy NVIDIA GB200 in data centers with lower temperatures, including Washington state in the US, Quebec City in Canada, Helsinki in Finland, etc to "proactively mitigate potential impacts from insufficient time to optimize the cooling system fully".

Kuo came to the following conclusions:

1. Microsoft's demand for GB200 surpasses the total orders from other CSPs, with 4Q24 orders significantly increasing by 3-4 times.
2. Microsoft's GB200 key component suppliers will begin mass production and shipment in 4Q24, contributing to their supply chain's performance earlier than other CSPs.
3. Revenue recognition for Microsoft's GB200 key component suppliers will precede the assembly supplier's, as key components will ship regardless of the assembly supplier's ability to meet Microsoft's 4Q24 demand.