CoreWeave has announced plans to deploy NVIDIA’s Rubin inference accelerator across its cloud platform in the second half of 2026. Rubin is designed to accelerate agentic AI applications, multi-modal reasoning and large-scale batch inference, enabling CoreWeave to support up to 50% more concurrent model runs per GPU rack compared with prior architectures. By being one of the first cloud providers to offer Rubin at scale, CoreWeave expects to reduce latency for real-time AI use cases by 30% and increase inference throughput by over 40% on popular transformer-based models. This integration aligns with CoreWeave’s strategy to capture growing demand from enterprises and research institutions for specialized AI compute, where the market for inference services is projected to exceed $15 billion by 2027. The company has already begun pilot deployments in its Atlanta and Frankfurt data centers, with full availability slated for Q3 2026.

D.A. Davidson analyst Gil Luria, formerly a noted skeptic of CoreWeave’s business model, upgraded the stock from Underperform to Neutral following the NVIDIA Rubin announcement and steady quarterly revenue growth. In his note, Luria cited a 60% year-over-year increase in AI compute bookings in Q4 2025 and the company’s success in onboarding three hyperscale customers for custom GPU racks. He also pointed to CoreWeave’s improvement in utilization rates, which rose from 72% to 85% over the past six months, and reaffirmed his forecast for 2026 revenue growth of approximately 80%. While Luria sees near-term capital intensity, he believes the accelerated rollout of Rubin-powered infrastructure will drive operating leverage and narrow the gap with larger cloud incumbents by late 2026.