GPUs are in big demand due to the AI Supercycle, and even five-year-old chips are seeing price bumps as NVIDIA CEO calls it "Good Wine".
NVIDIA's Fine Wine Moment: GPU Demand So High That Five-Year-Old Chips Are Rising In Prices
Fine Wine, a terminology that has long been associated with GPUs since the AMD Vega days, is now being used for NVIDIA's GPUs, and the reason is quite the opposite of how we think about Fine Wine when talking about chips.
Previously, the term was associated with GPUs when they saw a massive uplift in performance through optimizations to the driver stack, but now, it's being used for pricing.
As NVIDIA's CEO states, the demand for GPUs is explosive. Every data center in the world with AI needs is running GPUs, and while the CPU demand is going up drastically too, GPUs are still the primary compute component. This demand has led to a severe shortage of GPUs, leading semiconductor manufacturers to witness massive production and supply constraints.
This has resulted in a price hike across every single tech component, including GPUs. But it's not just new GPUs that are affected by this growing demand; older GPUs are just as much affected, and the demand is now causing a price hike across GPUs that are 4-5 years old.
NVIDIA's CEO states that the older GPUs are rising in price faster than good wine, meaning they are aging really well. The thing is that even though 4-5 years old refers to NVIDIA's Hopper generation of GPUs, the chips are still very much viable for AI and compute purposes thanks to continued optimizations, software-stack updates, and other rollouts.
Even Coreweave's CEO says that demand across older GPUs is accelerating, and prices of older GPUs such as the Hopper H100s, H200s, L40S, and even the generation prior, the A100s, are more expensive versus the previous quarter.
The company claims that they largely remain sold out on their capacity across the entire fleet, and that is something that we are seeing across all industry segments. Semiconductor makers are out of wafers, GPU/CPU/DRAM makers are out of each respective component, and AI Cloud providers are running out of compute capacity as demand for new AI models continues to accelerate.
Follow Wccftech on Google to get more of our news coverage in your feeds.
