NVIDIA's (NVDA, Financial) latest AI chip, Blackwell, is facing technical hurdles during data center deployment, including server rack overheating and chip connectivity issues. These problems have slowed deployment and prompted companies like Microsoft (MSFT), Amazon's AWS (AMZN), Google (GOOGL), and Meta (META) to cut orders for the Blackwell GB200 racks. As a result, Microsoft's Phoenix data center, initially intended for a large installation of GB200 chips, is now filled with NVIDIA's previous generation H200 chips.
If unresolved, these issues could lead to the Blackwell chip's performance falling short of promises. Following this news, NVIDIA's stock saw a drop of over 4% early in the session, later restraining to a 2.69% decline at $132.26 per share. Despite challenges, the Blackwell chip, notable for its performance and energy efficiency improvements over the prior Hopper line, continues to attract significant interest, with each tech giant placing orders exceeding $10 billion.
The integration of high-power chips into a single server rack has proven more challenging than expected. Each rack, towering over household refrigerators and weighing as much as a Honda Civic, requires a water cooling system due to high computational density, unlike traditional air cooling systems. Many data centers are not equipped to accommodate these racks, necessitating strategic deployment adjustments.
NVIDIA could potentially recover if technical issues are resolved and customers renew their interest in the Blackwell racks. Despite current setbacks, the chip remains superior to its predecessor, which may help NVIDIA find alternative buyers for the problematic racks.
