Nvidia’s Vera Rubin is months away — Blackwell is getting faster right now

Nvidia announced its next-generation Vera Rubin GPU, promising significant performance leaps in 2026, including 5x inference and 3.5x training gains over Blackwell. Despite the Vera Rubin announcement, Nvidia is also actively enhancing its current Blackwell architecture through ongoing optimizations. Blackwell, already in use, is seeing performance gains: up to 2.8x improvement in inference and 1.4x higher training performance. These improvements stem from innovations like programmatic dependent launch and optimized training recipes, leveraging NVFP4 precision. While Vera Rubin offers superior efficiency and cost savings for future deployments, Blackwell remains a viable platform for current and near-term AI initiatives. Enterprises can benefit from immediate Blackwell performance upgrades, increasing throughput and lowering costs. Investing in Blackwell in 2026 makes sense, but enterprises building large-scale infrastructure for 2026 should consider Vera Rubin. The strategic approach involves a phased deployment, maximizing Blackwell's value now while planning for Vera Rubin's benefits later. This approach ensures maximum value from current investments while staying competitive for the future.