Boost Run hits Blackwell Exemplar

Boost Run said on April 13 that it cleared NVIDIA’s Exemplar Cloud bar for Blackwell systems, a benchmark program for large artificial intelligence training runs. (prnewswire.com) NVIDIA describes Exemplar Cloud as a performance program for cloud providers, not a chip resale badge. The company says it measures workload performance, security, and reliability with standardized benchmarking recipes on production-scale infrastructure. (nvidia.com) For Boost Run, the timing is tied to a public-markets deal. The company said it has a definitive agreement to merge with Willow Lane Acquisition Corp., a special purpose acquisition company listed on Nasdaq under the ticker WLAC, and expects to trade as BRUN after closing. (prnewswire.com) Blackwell is NVIDIA’s latest data-center architecture for training and running large artificial intelligence models, the software behind chatbots, code generators, and image tools. NVIDIA says Blackwell systems are built for higher performance and lower energy use at very large scale. (nvidia.com) The cloud market around those chips has gotten more crowded. NVIDIA’s own cloud-partner directory lists a global roster of providers selling access to its accelerated-computing systems, while DGX Cloud Lepton is designed to route developers to outside GPU capacity across that network. (nvidia.com, nvidia.com) That has shifted the sales pitch from “we have graphics processing units” to “we can keep them busy efficiently.” NVIDIA says Exemplar Cloud is meant to give buyers predictable performance-per-cost data across providers using the same benchmark methods. (nvidia.com) Boost Run said Exemplar status puts it in a small group that includes CoreWeave, Nebius, Oracle Cloud Infrastructure, and Microsoft Azure. That peer set matters because those companies compete for the same training jobs from model builders and enterprises moving beyond experiments into full deployments. (tmcnet.com, prnewswire.com) NVIDIA has been pushing public benchmark results for Blackwell as a selling point. In November 2025, the company said Blackwell posted the fastest times across every MLPerf Training v5.1 benchmark, including training Llama 3.1 405B in 10 minutes on 5,120 Blackwell graphics processing units. (developer.nvidia.com) Boost Run’s announcement does not disclose prices, cluster size, or benchmark-by-benchmark results. What it does give the company is a vendor-backed performance credential at a moment when access to top-end artificial intelligence hardware is no longer rare, but proof that a cloud can run it well still is. (prnewswire.com, nvidia.com)