Nvidia CEO Jensen Huang compares the outdated costs of older AI graphics playing cards to advantageous wine – and CoreWeave confirms: Even older chips are getting costlier each quarter.
Nvidia CEO Jensen Huang has an uncommon comparability on the subject of the worth developments of older AI accelerators—as a result of, very like “our” gaming GPUs, costs are additionally rising on this sector.
Particularly, Huang said, based on »TheTranscript,” Huang commented as follows throughout an investor name:
GPU demand is skyrocketing—even chips we bought 4 or 5 years in the past are rising in worth sooner than advantageous wine.
AI Supercycle: Why Even Outdated Server GPUs Know No Relaxation
The time period “Positive Wine” has its personal historical past within the GPU world. Its roots lie within the AMD Vega period: Again then, it referred to graphics playing cards whose efficiency constantly improved over time by driver updates.
Nvidia is popping the tables, so to talk: With its AI accelerators, it’s not the efficiency that matures—however the worth.
CoreWeave additionally confirms the development. Based on the cloud supplier specializing in AI infrastructure, common costs for older Nvidia chips (together with the A100, H100, H200, and L40) have risen quarter-over-quarter.
$CRWV CoreWeave CEO: Pricing for older GPUs is rising
“Consequently, demand is accelerating throughout the board. Common pricing for the A100s, H100s, H200s, and L40s all elevated quarter-over-quarter, and we stay largely bought out for near-term capability throughout our fleet. This…
— The Transcript (@TheTranscript_) Might 11, 2026
The backdrop is the so-called “AI supercycle.” This refers back to the industry-wide, unprecedented demand for computing capability for synthetic intelligence.
In spite of everything, each information middle that makes use of AI in any method depends on such GPUs as its major computing unit—which is why even common high-end graphics playing cards are onerous to return by at “common” costs.On the identical time, the ensuing GPU scarcity is placing each producers and suppliers below stress to extend manufacturing and meet supply calls for. The result’s a shift towards older AI accelerators, which in flip are additionally turning into costlier.
No Finish to the Scarcity in Sight
The chips Huang refers to in his wine analogy are largely from the Hopper era—specifically the H100 and H200, accelerators developed 4 to 5 years in the past.
Regardless of their age, the older generations are nonetheless thought-about aggressive for AI and information middle workloads as a result of Nvidia constantly improves the chips by common driver and software program updates.
As well as, newer alternate options like Blackwell are nonetheless not obtainable in enough portions—very like what we’re presently seeing within the gaming market.
