CoreWeave shares jumped 8.4% on Wednesday after the cloud computing firm reported record-breaking performance results using Nvidia's latest Grace Blackwell chips. The shares surged more than 25% on Tuesday.

In partnership with Nvidia and IBM, CoreWeave ran 2,496 GB200 GPUs on its AI-focused cloud platform to complete the industry's largest-ever MLPerf Training v5.0 submission. The benchmarked cluster was over 30 times larger than that of the only other cloud provider participant.

The company said the system achieved a top result using the Llama 3.1 405B foundational model, finishing the run in just over 27 minutes. Training performance across similar cluster sizes was more than twice as fast compared to rival submissions, according to CoreWeave.

The firm said these performance gains are critical for speeding up AI development cycles and reducing training costs for clients. CoreWeave added that its customers may be able to deploy models months ahead of competitors using its infrastructure.

The company brought the GB200 GPUs online in April, following its public listing in March, and was the first cloud provider to offer them broadly at scale.