Wall Street may be obsessing over Nvidia Corp‘s (NASDAQ:NVDA) next-generation Blackwell chips, but the company’s partner, CoreWeave, Inc. (NASDAQ:CRWV), just hinted that an overlooked trade could be older GPUs.
Older-generation Nvidia GPUs — including Hopper and Ampere systems — are "going up across the board," CoreWeave co-founder Brannin McBee said at a J.P. Morgan TMC Conference this week.
AI customers increasingly choose chips based on workload economics rather than simply chasing the newest hardware, he explained.
That's an unusual dynamic in semiconductors, where aging chips typically lose value rapidly as newer generations arrive.
Instead, CoreWeave suggested the AI market is beginning to resemble infrastructure finance more than traditional hardware cycles.
Older Nvidia GPUs Still Printing Money
McBee pointed to older Tesla and Volta GPUs from the late 2010s that continue operating profitably across cloud environments despite being well beyond standard depreciation timelines.
"They’re not running for free and for fun. They’re running because they’re profitable," he said.
CoreWeave added that the industry-standard six-year depreciation life for GPUs "might end up being conservative," implying that AI compute assets may remain economically valuable far longer than investors initially expected.
The comments reinforce a growing view across the AI infrastructure trade that demand is broadening beyond massive frontier-model training runs.
While Nvidia's newest Blackwell systems remain the centerpiece for cutting-edge AI development, older Hopper and Ampere GPUs are increasingly finding roles in inference workloads, enterprise AI deployments and cost-sensitive compute tasks.
That shift could create a longer monetization tail for Nvidia's installed GPU base.
CoreWeave Throws Weight Behind Nvidia Ecosystem
CoreWeave also pushed back against growing speculation that AI customers may aggressively diversify into alternatives like Google TPUs or custom AI accelerators.
Addressing the recent Blackstone-Google TPU cloud partnership announcement, McBee said customer demand remains overwhelmingly centered around Nvidia GPUs.
"Our clients come to us asking us to build GPUs and explicitly NVIDIA GPUs," he said.
The company added that it remains "client led" and would consider alternative architectures only if customer demand reached sufficient scale.
The remarks could strengthen the bullish case that Nvidia's moat extends beyond raw chip performance into software compatibility, deployment familiarity and ecosystem lock-in.
For now, CoreWeave's message was clear: in the AI boom, even Nvidia's older GPUs are suddenly becoming prized assets again.
Photo Courtesy: Samuel Boivin on Shutterstock.com
Login to comment