Nvidia's Buck Says CoreWeave V100 GPUs Still Run After a Decade

Nvidia's vice president of hyperscale and HPC, Ian Buck, said CoreWeave is still running V100 GPUs nearly a decade after their release, according to a report relayed via Google News. The comment surfaced alongside news that Nvidia's Vera Rubin NVL72 platform has gone live, with the company claiming a 4.8x token-generation speedup over prior systems.

What Did Ian Buck Say About CoreWeave's V100 GPUs?

Buck's remark, as reported, points to V100 accelerators — chips Nvidia first shipped years ago — still operating in CoreWeave's cloud fleet. The report does not specify which workloads those V100s currently handle or how many units remain active, so those details cannot be confirmed here. The core claim is narrow but notable: silicon designed for an earlier generation of AI and HPC work is still in service at a major GPU cloud provider.

What Is Vera Rubin NVL72 and What Speedup Does Nvidia Claim?

The same report ties Buck's comment to the rollout of Vera Rubin NVL72, a newer Nvidia platform described as now live. Nvidia's associated claim is a 4.8x improvement in token-generation throughput compared with earlier hardware, per the report. No independent benchmark, methodology, or baseline configuration was included in the available source material, so the figure should be read as a company-attributed claim rather than a third-party-verified result.

Why Does GPU Longevity Matter for Cloud Providers?

For companies like CoreWeave that lease GPU capacity, hardware lifespan affects depreciation schedules, resale value, and how long older capacity can keep generating revenue before customers demand newer architectures. Buck's point, as framed in the report, underscores that even as Nvidia pushes new platforms with large generational speedups, older chips such as the V100 have not been fully retired from commercial use.

How Big a Jump Is 4.8x in Practical Terms?

The report frames the 4.8x figure specifically as a token-generation metric, which is commonly used to describe large language model inference speed. No additional context — such as model size, batch configuration, or power draw — was provided in the source material, limiting how far that number can be extrapolated to other workloads.

What's Next to Watch

• Whether CoreWeave or Nvidia publishes further detail on which V100 workloads remain active and for how long they will stay in service.
• Independent benchmarks of Vera Rubin NVL72 that could confirm or contextualize the 4.8x token-generation claim.
• Broader industry commentary on GPU depreciation cycles as new platforms arrive faster than older hardware is retired.

Figures Cited

• Nearly a decade: approximate age of the V100 GPUs still reported running at CoreWeave, per Ian Buck's comments.
• 4.8x: the token-generation speedup Nvidia attributes to the newly live Vera Rubin NVL72 platform, per the report.

The original report is available via Google News.