CoreWeave Sets New Inference Performance Benchmarks with NVIDIA's Latest AI Hardware
Event summary
- CoreWeave achieved leading inference performance in MLPerf® Inference v6.0 using NVIDIA’s GB200 NVL72 and GB300 NVL72.
- The company demonstrated standout throughput on DeepSeek-R1’s sparse Mixture-of-Experts architecture.
- CoreWeave's results showed 2X improvement over its own MLPerf® 5.1 results on the same hardware footprint.
- Eight of the top 10 model providers rely on CoreWeave Cloud for their AI workloads.
The big picture
CoreWeave's MLPerf v6.0 results underscore the strategic shift towards inference as the critical measure of AI performance. As enterprises move from experimentation to production, the gap between theoretical and real-world output is becoming a defining constraint. CoreWeave’s ability to close this gap through full-stack optimization positions it as a key player in the AI infrastructure landscape.
What we're watching
- Performance Scaling
- How CoreWeave will sustain its performance lead as AI inference demands grow.
- Market Differentiation
- Whether CoreWeave's full-stack optimization can maintain its competitive edge in the AI cloud market.
- Customer Adoption
- The pace at which leading model providers will migrate to CoreWeave’s platform for production workloads.