TLDR
- CRWV stock gains as CoreWeave posts record DeepSeek training run
- CoreWeave trains DeepSeek-V3 671B in just 2.02 minutes
- CRWV uses 8,192 NVIDIA GB300 GPUs for benchmark record
- MLPerf results highlight CoreWeave’s large AI cloud scale
- CoreWeave says production cloud powered its fastest benchmark run
CoreWeave stock traded at $117.68, up 10.28%, after the company reported record MLPerf Training v6.0 results. The AI cloud firm trained DeepSeek-V3 671B in 2.02 minutes. The result strengthened CRWV’s profile in high-performance AI infrastructure.
CoreWeave, Inc. Class A Common Stock, CRWV
CoreWeave Sets DeepSeek Training Benchmark
Coreweave used 8,192 NVIDIA GB300 NVL72 GPUs across 2,048 nodes for the record run. The company said the setup marked the largest GB300 cluster submitted in the benchmark round. It also used the same cloud infrastructure available to customers.
The company submitted three GB300 NVL72 configurations for DeepSeek-V3 671B. Its 4,096-GPU setup completed training in 3.09 minutes. The 2,048-GPU setup completed the workload in 5.54 minutes.
The results showed steady scaling as CoreWeave doubled cluster size. The company highlighted full-stack performance rather than raw GPU count alone. The benchmark also placed CoreWeave ahead across closed and available-cloud submissions.
GB300 Scale Strengthens CRWV AI Cloud Case
CoreWeave said the benchmark showed the value of networking, scheduling, storage, and orchestration. These systems support large workloads and reduce delays across major training clusters. The company framed the result as a production cloud milestone.
The company also tested Llama-3.1-405B on a 4,096-GPU GB300 deployment. That run reached the reference quality target in 9.77 minutes. CoreWeave said the result used 20% fewer GPUs than larger GB200 deployments.
CoreWeave also reported smaller-cluster results using NVIDIA HGX B200 systems. It trained GPT-OSS-20B in 26.98 minutes on a 64-GPU cluster. It also trained Llama-3.1-8B in 16.54 minutes on the same setup.
Production Infrastructure Supports Benchmark Results
CoreWeave said its Mission Control platform helped maintain fleet-wide performance consistency. The system checks hardware, firmware, network, and thermal health during large training jobs. This process helps reduce slow nodes and keeps workloads stable.
The company also used topology-aware scheduling through CoreWeave SUNK. That system places workloads to improve locality within NVL72 domains. In addition, rail-aware networking balances traffic and helps prevent bottlenecks at large scale.
CoreWeave completed its Nasdaq listing in March 2025. The company now serves AI labs, startups, and enterprise customers with cloud infrastructure. Its latest benchmark gives CRWV another data point in the race for faster model training.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
