tech
Why soaring memory prices could be Nvidia's next big advantage
Nvidia, led by CEO Jensen Huang, faces rising memory costs impacting AI infrastructure and partnerships on new systems.
TL;DR
- Rising memory prices are increasing the cost of AI infrastructure.
- Meta and Microsoft have cited rising memory costs as a reason for increased AI spending.
- Nvidia's upcoming Vera Rubin NVL72 system is estimated to cost $9.1 million per rack.
- High-bandwidth memory prices are projected to more than triple, pushing memory and storage costs to over a third of the total rack price.
- A 1-gigawatt Vera Rubin data center could cost $47 billion to build, an increase from predecessor chips.
- Networking, cooling, and power-delivery costs are also rising.
- Nvidia is expected to pass on higher memory costs to customers due to its strong market position.
- Nvidia CEO Jensen Huang has warned that memory shortages could last for years.
- Nvidia and SK Hynix have partnered to develop next-generation memory for future systems.
- SK Hynix recently reached a market capitalization of $1 trillion.