Nvidia launches Nemotron 3.5 Super VL and Bolt for AI agents - Techzine Global

Nvidia's bundling of Nemotron 3.5 Super VL with Bolt into the CUDA stack is a strategic pivot from compute vendor to AI operating-system layer. The technical ripple is immediate: multimodal inference throughput demands will stretch CoWoS packaging capacity and HBM3E order cycles, while Bolt's agent-orchestration layer compresses latency to sub-millisecond levels, triggering a structural jump in Jetson edge shipments by H2 2025. On compliance, agent frameworks touching cross-border data flows and autonomous decision chains will face granular scrutiny under the EU AI Act and U.S. export controls. TSMC's exclusive CoWoS-L capacity in Taiwan, China remains both a moat and a single-point-of-failure risk. Competitively, AMD's MI350 plus ROCm trails by roughly two iteration cycles in agent-ecosystem readiness. Huawei's Ascend 910C and MindSpore build a parallel domestic track, yet the depth gap in multimodal operator libraries won't close within two quarters. My 12-to-24-month call: Nvidia's software-and-model revenue share will climb from under 5% past 15%, re-rating the stock from a cyclical chip play to an AI platform franchise. Agent-scale deployment will replicate the iPhone moment for mobile apps—Nvidia is positioning itself as the iOS of autonomous intelligence.