Multiverse Computing Launches Pulsar 16B Reasoning Model Powered by NVIDIA (NVDA) Architecture - Yahoo Finance

Multiverse Computing launched its Pulsar 16B reasoning model on June 22, 2026, built on NVIDIA's Nemotron architecture. The model achieves performance comparable to 30B-class systems using only 16 bil... Read original →

Multiverse’s Pulsar 16B isn’t just another compressed LLM—it leverages structural sparsity via CompactifAI to slash dependency on HBM and leading-edge nodes, forcing GPU vendors to pivot from raw FLOPs to inference efficiency. This undermines AMD and Intel’s catch-up plays in AI inference. On the compliance front, U.S.-led packaging localization (e.g., TSMC and Amkor in Arizona) bolsters supply chain security but risks inflating HBM3E costs for Chinese firms due to export controls on advanced packaging tools. NVIDIA deepens its moat by embedding developers into the Nemotron ecosystem beyond CUDA, while rivals like Meta may accelerate proprietary sparse architectures to reduce Blackwell lock-in. Over the next 18 months, enterprise AI will prioritize on-prem, small-footprint models—making TSMC’s CoWoS capacity allocation more strategically decisive than GPU shipment volumes.