NVIDIA lines up $500B AI financing and launches Nemotron 3.5 Lightning

NVIDIA moved on two fronts. On financing, it signed memoranda of understanding with six major financial institutions — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to mobilize more than $500 billion for AI infrastructure buildout, a Reuters Breakingviews piece framing Jensen Huang as taking the wheel of a '$500 billion AI bandwagon.' The scale underscores how compute financing itself has become a strategic battleground.

On product, NVIDIA released Nemotron 3.5 Lightning, a 30-billion-parameter open mixture-of-experts model with roughly 3 billion active parameters that it says delivers 4x the throughput for always-on agents. Distributed under the OpenMDW license and supporting Blackwell, Hopper and Ampere, it features a Mamba-2/MoE/Attention hybrid architecture with up to a 1M-token context, aimed squarely at long-running autonomous agents and efficient local inference. Alongside it, NVIDIA shipped NeMo Switchyard, an open-source routing library that picks which model runs for a given agent task — early adopters include CrowdStrike, Harvey and CodeRabbit.

The positioning is telling: with Meta's Muse Glimmer and Alibaba's Qwen line, NVIDIA is helping seed the open 30B local-agent tier while its Switchyard router quietly aims to sit above the whole ecosystem deciding 'who decides which model runs.' A reported 1-trillion-parameter Nemotron 4 is said to be due late fall 2026. The through-line — from half-a-trillion in financing to an agent-routing layer — is NVIDIA extending beyond silicon into the economics and orchestration of the AI stack.